[{"data":1,"prerenderedAt":5155},["ShallowReactive",2],{"header-counts":3,"footer-counts":6,"tool-\u002Ftools\u002Fcoding\u002Flocal\u002Fjan":8,"cat-rank-coding-local":506,"tool-related-coding\u002Flocal\u002Fjan":4112,"tool-reviews-coding\u002Flocal\u002Fjan":4113,"tool-alts-coding\u002Flocal\u002Fjan":4114},{"tools":4,"reviews":5},98,27,{"tools":4,"reviews":5,"playbooks":7,"news":5},22,{"id":9,"title":10,"alternatives":11,"api_compatible":15,"body":16,"category":470,"chinese_friendly":456,"cover":471,"description":472,"domestic":473,"extension":474,"faq":15,"free":473,"github":451,"languages":475,"lastVerified":477,"meta":478,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":480,"pillar":481,"platforms":482,"priceTable":15,"pricing":486,"published":487,"relatedPlaybooks":15,"relatedReviews":15,"score":488,"self_host":473,"seo":491,"seoTitle":492,"slug":493,"sources":494,"stem":497,"suitable":15,"tagline":498,"tags":499,"updated":477,"verdict":504,"website":443,"__hash__":505},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fjan.md","Jan",[12,13,14],"coding\u002Flocal\u002Flm-studio","coding\u002Flocal\u002Fcherry-studio","coding\u002Flocal\u002Follama",null,{"type":17,"value":18,"toc":454},"minimark",[19,24,33,36,39,91,94,100,133,136,140,148,165,170,187,190,211,214,326,329,361,365,388,392,398,404,410,413,429,432,437],[20,21,23],"h2",{"id":22},"tldr","TL;DR",[25,26,27,28,32],"p",{},"Jan 是开源的本地 LLM 桌面客户端，定位是",[29,30,31],"strong",{},"ChatGPT 的离线替代品","。界面设计精美，操作体验接近 ChatGPT，支持 GGUF 模型一键下载、本地推理、多模型切换、插件扩展。AGPL 开源，完全免费。适合想要一个好看好用的本地 AI 聊天工具、隐私优先的用户。",[25,34,35],{},"适合：想要 ChatGPT 颜值和体验的本地替代、个人离线聊天、隐私敏感场景、非技术用户（GUI 友好）。不适合：需要 OpenAI 兼容 API 给应用接入（用 Ollama）、需要模型调参 \u002F 量化选择（用 LM Studio）、企业商用（AGPL 限制）。",[20,37,38],{"id":38},"核心能力",[40,41,42,49,55,61,67,73,79,85],"ul",{},[43,44,45,48],"li",{},[29,46,47],{},"ChatGPT 式界面","：聊天 UI 设计精美，多会话管理、Markdown 渲染、代码高亮",[43,50,51,54],{},[29,52,53],{},"一键下载模型","：内置模型市场，搜索 GGUF 模型点击下载，自动配置",[43,56,57,60],{},[29,58,59],{},"本地推理","：基于 llama.cpp，支持 CPU \u002F GPU 加速，完全离线运行",[43,62,63,66],{},[29,64,65],{},"多模型切换","：一个会话可切换不同模型对比输出，方便评估",[43,68,69,72],{},[29,70,71],{},"插件系统","：支持扩展功能，如网页搜索、文档分析、API 代理等",[43,74,75,78],{},[29,76,77],{},"远程 API 接入","：除了本地模型，也支持接 OpenAI \u002F Anthropic 等云端 API",[43,80,81,84],{},[29,82,83],{},"跨平台桌面 App","：Win \u002F Mac \u002F Linux 原生安装包，Electron 构建",[43,86,87,90],{},[29,88,89],{},"隐私优先","：所有数据本地存储，无遥测，无云端调用（本地模型模式）",[20,92,93],{"id":93},"价格",[95,96,97],"blockquote",{},[25,98,99],{},"以下信息为 2026-07-30 核实。",[101,102,103,118],"table",{},[104,105,106],"thead",{},[107,108,109,113,115],"tr",{},[110,111,112],"th",{},"方案",[110,114,93],{},[110,116,117],{},"说明",[119,120,121],"tbody",{},[107,122,123,127,130],{},[124,125,126],"td",{},"开源版",[124,128,129],{},"$0",[124,131,132],{},"完整功能，AGPL 协议",[25,134,135],{},"完全免费。注意 AGPL 协议：个人使用无限制，但二次开发 \u002F 商用需遵守开源传染条款。",[20,137,139],{"id":138},"体验与评测资料整理","体验与评测（资料整理）",[95,141,142],{},[25,143,144,145],{},"说明：本节基于官方文档与公开评测整理，非本站独立实测环境，具体数据请以官方为准。\n",[29,146,147],{},"亮点：",[40,149,150,153,156,159,162],{},[43,151,152],{},"界面设计是同类最佳：比 LM Studio \u002F GPT4All 好看很多，接近 ChatGPT 体验",[43,154,155],{},"模型下载体验顺滑：搜索 → 下载 → 使用，全程 GUI，零命令行",[43,157,158],{},"多模型对比实用：同一问题切换模型看不同回答，选模型很方便",[43,160,161],{},"插件系统有潜力：网页搜索插件让本地模型也能联网",[43,163,164],{},"支持云端 API 混用：本地模型 + GPT-4o 切换，一个客户端搞定",[25,166,167],{},[29,168,169],{},"踩坑：",[40,171,172,175,178,181,184],{},[43,173,174],{},"Electron 应用内存占用偏高，老设备偶有卡顿",[43,176,177],{},"模型管理不如 LM Studio：量化版本选择少，调参选项有限",[43,179,180],{},"API Server 功能弱：有 OpenAI 兼容端点但不如 Ollama 灵活",[43,182,183],{},"插件生态尚不成熟，可用插件不多",[43,185,186],{},"AGPL 协议对企业不友好，商用需注意合规",[20,188,189],{"id":189},"上手",[191,192,193,196,199,202,205,208],"ol",{},[43,194,195],{},"从 jan.ai 下载对应平台安装包",[43,197,198],{},"安装后打开 Jan，界面类似 ChatGPT",[43,200,201],{},"点击模型市场（Hub）→ 搜索推荐模型（Qwen2.5-7B \u002F Llama3.1-8B）",[43,203,204],{},"下载模型后，新建会话 → 选择模型 → 开始聊天",[43,206,207],{},"多模型对比：同一会话切换模型或开多个会话",[43,209,210],{},"接云端 API：Settings → API Keys → 填入 OpenAI Key 即可混用",[20,212,213],{"id":213},"对比",[101,215,216,234],{},[104,217,218],{},[107,219,220,223,225,228,231],{},[110,221,222],{},"维度",[110,224,10],{},[110,226,227],{},"LM Studio",[110,229,230],{},"Ollama",[110,232,233],{},"Cherry Studio",[119,235,236,252,266,281,295,309],{},[107,237,238,241,244,247,250],{},[124,239,240],{},"界面颜值",[124,242,243],{},"高",[124,245,246],{},"中",[124,248,249],{},"无 GUI",[124,251,243],{},[107,253,254,257,259,262,264],{},[124,255,256],{},"模型管理",[124,258,246],{},[124,260,261],{},"强",[124,263,261],{},[124,265,246],{},[107,267,268,271,274,277,279],{},[124,269,270],{},"API 接入",[124,272,273],{},"基础",[124,275,276],{},"✅",[124,278,261],{},[124,280,276],{},[107,282,283,286,288,291,293],{},[124,284,285],{},"插件扩展",[124,287,276],{},[124,289,290],{},"❌",[124,292,290],{},[124,294,276],{},[107,296,297,300,302,304,307],{},[124,298,299],{},"云端 API 混用",[124,301,276],{},[124,303,276],{},[124,305,306],{},"需配",[124,308,276],{},[107,310,311,314,317,320,323],{},[124,312,313],{},"开源协议",[124,315,316],{},"AGPL",[124,318,319],{},"闭源",[124,321,322],{},"MIT",[124,324,325],{},"Apache",[20,327,328],{"id":328},"避坑",[40,330,331,337,343,349,355],{},[43,332,333,336],{},[29,334,335],{},"别指望它做 API 服务器","：Jan 的 API Server 功能基础，给应用接入用 Ollama",[43,338,339,342],{},[29,340,341],{},"模型选对量化","：默认下载的可能不是最优量化，手动选 Q4_K_M 平衡速度质量",[43,344,345,348],{},[29,346,347],{},"Electron 吃内存","：8GB RAM 设备跑大模型 + Jan 本身会卡，关其他应用",[43,350,351,354],{},[29,352,353],{},"AGPL 商用注意","：企业内部署需法务确认 AGPL 合规",[43,356,357,360],{},[29,358,359],{},"插件别装太多","：部分插件质量参差，可能影响稳定性",[20,362,364],{"id":363},"适合-不适合","适合 \u002F 不适合",[40,366,367,370,373,376,379,382,385],{},[43,368,369],{},"✅ 想要 ChatGPT 颜值和体验的本地替代",[43,371,372],{},"✅ 个人离线聊天 \u002F 隐私优先场景",[43,374,375],{},"✅ 非技术用户（GUI 友好，零命令行）",[43,377,378],{},"✅ 本地 + 云端 API 混用需求",[43,380,381],{},"❌ 需要给应用 \u002F IDE 接入 API（用 Ollama）",[43,383,384],{},"❌ 需要精细模型调参 \u002F 量化管理（用 LM Studio）",[43,386,387],{},"❌ 企业商用（AGPL 限制）",[20,389,391],{"id":390},"faq","FAQ",[25,393,394,397],{},[29,395,396],{},"Q: Jan 和 LM Studio 怎么选？","\nA: 颜值和聊天体验选 Jan，模型管理和调参选 LM Studio。Jan 更像 ChatGPT，LM Studio 更像模型工具箱。两者都免费，可以都装。",[25,399,400,403],{},[29,401,402],{},"Q: 能给 Cursor \u002F Cline 接入吗？","\nA: Jan 有 OpenAI 兼容 API Server（默认端口 1337），理论上可以。但不如 Ollama 稳定灵活，推荐用 Ollama 做 API 服务器。",[25,405,406,409],{},[29,407,408],{},"Q: AGPL 协议影响个人使用吗？","\nA: 不影响。AGPL 主要约束网络服务分发场景。个人本地使用完全无限制。只有你把 Jan 改造后对外提供 SaaS 服务才需开源你的修改。",[20,411,412],{"id":412},"相关阅读",[25,414,415,420,421,420,425],{},[416,417,419],"a",{"href":418},"\u002Fcoding\u002Flocal\u002Fgpt4all.html","GPT4All"," · ",[416,422,424],{"href":423},"\u002Fcoding\u002Flocal\u002Fvllm.html","vLLM",[416,426,428],{"href":427},"\u002Fagent\u002Fdesktop\u002Fopen-interpreter.html","Open Interpreter",[20,430,431],{"id":431},"来源",[95,433,434],{},[25,435,436],{},"本文的价格、版本号与性能数据均参考以下官方渠道整理，可能随时间变动，请以官方实时信息为准。",[40,438,439,447],{},[43,440,441],{},[416,442,446],{"href":443,"rel":444},"https:\u002F\u002Fjan.ai",[445],"nofollow","官网",[43,448,449],{},[416,450,453],{"href":451,"rel":452},"https:\u002F\u002Fgithub.com\u002Fjanhq\u002Fjan",[445],"GitHub",{"title":455,"searchDepth":456,"depth":456,"links":457},"",3,[458,460,461,462,463,464,465,466,467,468,469],{"id":22,"depth":459,"text":23},2,{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":138,"depth":459,"text":139},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":390,"depth":459,"text":391},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"local","\u002Fimg\u002Ftools\u002Fjan.webp","Jan 真实评测：开源本地 LLM 桌面客户端（AGPL 协议），定位 ChatGPT 的离线替代，支持 GGUF 模型一键下载 + 本地推理 + 插件扩展。跨平台桌面 app，适合需要完全离线、隐私优先的本地 AI 聊天场景。",false,"md",[476],"en","2026-07-30",{},true,"\u002Ftools\u002Fcoding\u002Flocal\u002Fjan","coding",[483,484,485],"windows","macos","linux","Free \u002F 开源（AGPL）","2026-07-05",{"power":456,"ux":489,"price":490,"cn_support":456,"stability":456},4,5,{"title":10,"description":472},"Jan - 开源本地 LLM 桌面客户端评测 | AIHO","coding\u002Flocal\u002Fjan",[495,496],{"title":446,"url":443},{"title":453,"url":451},"tools\u002Fcoding\u002Flocal\u002Fjan","开源本地 LLM 桌面客户端，定位 ChatGPT 的离线替代",[470,500,501,502,503],"desktop","opensource","gguf","offline","颜值最高、最像 ChatGPT 的开源本地 LLM 客户端，离线聊天体验好；但 API 能力和模型管理不如 Ollama\u002FLM Studio，定位偏轻量个人使用。","SYftPw58WAWIAEA4qHpFo2Ki3Z1H9q97Rp4jQR560KE",[507,983,1390,1718,2195,2682,3145,3610],{"id":508,"title":233,"alternatives":509,"api_compatible":15,"body":512,"category":470,"chinese_friendly":490,"cover":926,"description":927,"domestic":473,"extension":474,"faq":928,"free":473,"github":15,"languages":941,"lastVerified":15,"meta":943,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":944,"pillar":481,"platforms":945,"priceTable":947,"pricing":956,"published":957,"relatedPlaybooks":958,"relatedReviews":15,"score":961,"self_host":479,"seo":962,"seoTitle":963,"slug":13,"sources":964,"stem":972,"suitable":15,"tagline":973,"tags":974,"updated":967,"verdict":980,"website":981,"__hash__":982},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio.md",[510,12,14,511],"coding\u002Flocal\u002Flobe-chat","coding\u002Flocal\u002Fopen-webui",{"type":17,"value":513,"toc":914},[514,516,519,522,524,568,570,583,588,592,596,613,617,634,636,660,662,802,804,836,838,861,863,889,891],[20,515,23],{"id":22},[25,517,518],{},"Cherry Studio 是一款开源、跨平台（Windows \u002F macOS \u002F Linux \u002F Android）的桌面 AI 客户端，定位『全能 AI 工作台』：把 OpenAI \u002F Anthropic \u002F Google \u002F DeepSeek 等云端模型，以及 Ollama \u002F LM Studio 本地模型，全部聚合到同一个桌面应用里管理。内置 300+ 助手模板、本地 RAG 知识库、Markdown + Mermaid 渲染、MCP 协议支持，所有对话数据本地存储 + WebDAV 备份。AGPL-3.0 开源、GitHub 60k+ stars，企业版可联系商务做私有化部署。",[25,520,521],{},"适合：中文 AI 重度用户、想统一管理多家模型、需要本地知识库 RAG、关注数据本地存储的开发者 \u002F 研究者。不适合：要 Web 端访问 \u002F Docker 自托管 \u002F 团队多人共享 \u002F iOS 端使用。",[20,523,38],{"id":38},[40,525,526,532,538,544,550,556,562],{},[43,527,528,531],{},[29,529,530],{},"多模型聚合","：OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F Moonshot 等云端 + Ollama \u002F LM Studio 本地",[43,533,534,537],{},[29,535,536],{},"本地 RAG 知识库","：拖拽 PDF \u002F Word \u002F Excel \u002F PPT \u002F 网址 \u002F sitemap → 自动向量化 → 检索增强问答 + 来源追溯",[43,539,540,543],{},[29,541,542],{},"300+ 助手模板","：编程 \u002F 写作 \u002F 翻译 \u002F 学习 \u002F 角色扮演开箱即用，可自定义 System Prompt",[43,545,546,549],{},[29,547,548],{},"MCP 协议","：扩展工具调用 \u002F 联网搜索 \u002F 文件操作",[43,551,552,555],{},[29,553,554],{},"数据本地优先","：对话历史本地存储，WebDAV 同步，不上传第三方",[43,557,558,561],{},[29,559,560],{},"多模态","：图片识别 \u002F PDF 阅读 \u002F Markdown + Mermaid + 代码高亮",[43,563,564,567],{},[29,565,566],{},"AI 绘画 + 翻译","：内置主流 SD \u002F DALL·E \u002F 翻译 API 集成",[20,569,93],{"id":93},[40,571,572,577],{},[43,573,574,576],{},[29,575,126],{},"：完全免费，AGPL-3.0",[43,578,579,582],{},[29,580,581],{},"Enterprise","：私有化部署 + 团队协作 + 资源管控，联系销售",[95,584,585],{},[25,586,587],{},"模型 API 费用按你自己绑定的供应商计费；本地 Ollama \u002F LM Studio 零成本。",[20,589,591],{"id":590},"实测mac-m2-中型知识库","实测（Mac M2 + 中型知识库）",[25,593,594],{},[29,595,147],{},[40,597,598,601,604,607,610],{},[43,599,600],{},"中文 UI \u002F 文档 \u002F 社区都顶级，零门槛上手",[43,602,603],{},"本地 RAG 拖入 30+ PDF 后向量化 \u003C 2 分钟（用 bge-m3）",[43,605,606],{},"多模型并排回答：让 Claude \u002F GPT \u002F DeepSeek 同回一个问题做比较",[43,608,609],{},"MCP 接 Brave Search + 自定义工具流畅",[43,611,612],{},"WebDAV 同步坚果云 \u002F 阿里云盘，桌面 + 移动设备数据互通",[25,614,615],{},[29,616,169],{},[40,618,619,622,625,628,631],{},[43,620,621],{},"没有 Web 端 \u002F Docker 自托管（要这个用 LobeChat）",[43,623,624],{},"iOS 版尚未发布（roadmap 中）",[43,626,627],{},"大型 PDF（>100 MB）向量化偶有失败，要切小",[43,629,630],{},"助手市场质量参差，要自筛",[43,632,633],{},"模型 API 调用全靠你自己付费，新手要先理解 API Key 概念",[20,635,189],{"id":189},[191,637,638,641,644,651,654,657],{},[43,639,640],{},"cherry-ai.com 下载客户端（或 GitHub releases）",[43,642,643],{},"设置 → 模型服务 → 填 OpenAI \u002F Claude \u002F DeepSeek API Key",[43,645,646,647],{},"（可选）本地：装 Ollama → Cherry Studio 自动识别 endpoint ",[416,648,649],{"href":649,"rel":650},"http:\u002F\u002Flocalhost:11434",[445],[43,652,653],{},"新建知识库 → 拖文件 \u002F 加网址 → 等向量化",[43,655,656],{},"新对话 → 选模型 → 勾知识库 → 提问",[43,658,659],{},"进阶：自定义助手（System Prompt）+ MCP 扩展工具",[20,661,213],{"id":213},[101,663,664,680],{},[104,665,666],{},[107,667,668,670,672,675,677],{},[110,669,222],{},[110,671,233],{},[110,673,674],{},"LobeChat",[110,676,227],{},[110,678,679],{},"Open WebUI",[119,681,682,698,712,727,740,756,771,785],{},[107,683,684,687,690,693,695],{},[124,685,686],{},"形态",[124,688,689],{},"桌面",[124,691,692],{},"Web + 桌面",[124,694,689],{},[124,696,697],{},"Docker \u002F 桌面",[107,699,700,702,705,707,710],{},[124,701,530],{},[124,703,704],{},"✅ 云 + 本地",[124,706,704],{},[124,708,709],{},"本地为主",[124,711,704],{},[107,713,714,717,720,722,725],{},[124,715,716],{},"知识库 RAG",[124,718,719],{},"✅ 强",[124,721,719],{},[124,723,724],{},"弱",[124,726,276],{},[107,728,729,732,734,736,738],{},[124,730,731],{},"MCP",[124,733,276],{},[124,735,276],{},[124,737,724],{},[124,739,276],{},[107,741,742,745,748,751,754],{},[124,743,744],{},"自托管 \u002F Web",[124,746,747],{},"无 Web",[124,749,750],{},"✅ Docker",[124,752,753],{},"无",[124,755,750],{},[107,757,758,761,764,766,769],{},[124,759,760],{},"中文",[124,762,763],{},"5\u002F5",[124,765,763],{},[124,767,768],{},"4\u002F5",[124,770,768],{},[107,772,773,775,778,780,783],{},[124,774,313],{},[124,776,777],{},"AGPL-3.0",[124,779,322],{},[124,781,782],{},"闭源（免费）",[124,784,322],{},[107,786,787,790,793,796,799],{},[124,788,789],{},"GitHub Stars",[124,791,792],{},"60k+",[124,794,795],{},"72k+",[124,797,798],{},"–",[124,800,801],{},"126k+",[20,803,328],{"id":328},[40,805,806,812,818,824,830],{},[43,807,808,811],{},[29,809,810],{},"API Key 别明文外泄","：客户端配置文件以明文存 Key，机器借出前先清；团队共享用企业版 \u002F 自建中转",[43,813,814,817],{},[29,815,816],{},"知识库别一次塞太多","：单库 1000+ 文档检索质量明显下降，按主题切分多个知识库",[43,819,820,823],{},[29,821,822],{},"嵌入模型选择","：免费 bge-m3 够用；专业用付费 Pro\u002FBAAI\u002Fbge-m3 或 OpenAI text-embedding-3",[43,825,826,829],{},[29,827,828],{},"WebDAV 同步先小范围测","：知识库向量数据较大，先备份对话再开同步",[43,831,832,835],{},[29,833,834],{},"MCP 工具来源要可控","：MCP 是给 AI 真实工具能力，第三方插件审一遍代码",[20,837,364],{"id":363},[40,839,840,843,846,849,852,855,858],{},[43,841,842],{},"✅ 中文用户、AI 重度使用 \u002F 多模型管理",[43,844,845],{},"✅ 需要本地 RAG 知识库",[43,847,848],{},"✅ 关注数据隐私 \u002F 本地存储",[43,850,851],{},"✅ 想用 Ollama \u002F LM Studio 本地模型",[43,853,854],{},"❌ 需要 Web 端 \u002F Docker 自托管",[43,856,857],{},"❌ 团队多人共享 \u002F SSO",[43,859,860],{},"❌ iOS 主力用户",[20,862,412],{"id":412},[40,864,865,871,877,883],{},[43,866,867],{},[416,868,870],{"href":869},"\u002Ftools\u002Fcoding\u002Flocal\u002Flobe-chat","LobeChat 评测",[43,872,873],{},[416,874,876],{"href":875},"\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio","LM Studio 评测",[43,878,879],{},[416,880,882],{"href":881},"\u002Ftools\u002Fcoding\u002Flocal\u002Follama","Ollama 评测",[43,884,885],{},[416,886,888],{"href":887},"\u002Fplaybook\u002Fonboarding\u002Frag-pipeline-build","RAG Pipeline 搭建 Playbook",[20,890,431],{"id":431},[191,892,893,900,907],{},[43,894,895,896],{},"Cherry Studio 官网（功能 + 下载）",[416,897,898],{"href":898,"rel":899},"https:\u002F\u002Fwww.cherry-ai.com\u002F",[445],[43,901,902,903],{},"MBLUO Studio — Cherry Studio 评测 2026 ",[416,904,905],{"href":905,"rel":906},"https:\u002F\u002Fmbluostudio.com\u002Ftools\u002Fcherry-studio",[445],[43,908,909,910],{},"Cursor IDE 博客 — Cherry Studio 完全指南（2025-03）",[416,911,912],{"href":912,"rel":913},"https:\u002F\u002Fwww.cursor-ide.com\u002Fblog\u002Fcherry-studio-guide",[445],{"title":455,"searchDepth":456,"depth":456,"links":915},[916,917,918,919,920,921,922,923,924,925],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":590,"depth":459,"text":591},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Fcherry-studio.webp","Cherry Studio 真实评测：开源跨平台桌面 AI 客户端，集成 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek + Ollama \u002F LM Studio 本地模型，内置 300+ 助手模板 + 本地 RAG 知识库。AGPL-3.0 开源、GitHub 60k+ stars，企业版另询。",[929,932,935,938],{"q":930,"a":931},"Cherry Studio 真的免费吗？","是。客户端完全免费、AGPL-3.0 开源，模型调用走你自己的 API Key（OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek 等付费）或本地 Ollama \u002F LM Studio（零成本）。",{"q":933,"a":934},"本地知识库怎么用？","在『知识库』面板新建，拖文件 \u002F 加网址 \u002F 填 sitemap，系统自动向量化（默认 BAAI\u002Fbge-m3 或硅基流动的 Pro 版）；提问时勾选要检索的知识库，AI 会基于检索片段答题并标出来源。",{"q":936,"a":937},"和 LobeChat 怎么选？","都开源、多模型、有 RAG。LobeChat 是 Web + 桌面双形态，可自托管 Docker，72k stars；Cherry Studio 是纯桌面（Win\u002FMac\u002FLinux\u002FAndroid），不支持 Web 部署但桌面体验更精细，60k+ stars。要 Web 访问 \u002F 公司多人共享选 LobeChat；个人重度选 Cherry Studio。",{"q":939,"a":940},"支持 MCP \u002F 插件吗？","支持 MCP（Model Context Protocol）扩展，配合自定义助手（System Prompt）可扩展工具调用、联网搜索等能力。",[942,476],"zh",{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fcherry-studio",[483,484,485,946],"android",[948,952],{"plan":126,"price":949,"features":950,"notes":951},"免费","300+ 助手模板 \u002F 云端 + 本地模型 \u002F 知识库 \u002F MCP \u002F WebDAV 备份","AGPL-3.0 开源",{"plan":581,"price":953,"features":954,"notes":955},"联系销售","私有化部署 \u002F 团队协作 \u002F AI 资源管控 \u002F 知识库管理","面向企业团队","开源免费 \u002F 企业版联系销售","2026-06-19",[959,960],"onboarding\u002Frag-pipeline-build","onboarding\u002Fcursor-mcp-deep-integration",{"power":489,"ux":490,"price":490,"cn_support":490,"stability":489},{"title":233,"description":927},"Cherry Studio 评测 2026：AI 客户端工具，多模型桌面助手，开源免费",[965,968,970],{"name":966,"url":898,"accessed":967},"Cherry Studio 官网","2026-06-24",{"name":969,"url":905,"accessed":967},"MBLUO Studio — Cherry Studio 评测",{"name":971,"url":912,"accessed":967},"Cursor IDE 博客 — Cherry Studio 指南","tools\u002Fcoding\u002Flocal\u002Fcherry-studio","全能 AI 客户端：多模型聚合 + 本地知识库 + 300+ 助手模板，跨平台桌面应用",[470,500,975,976,977,978,979],"multi-model","knowledge-base","rag","open-source","china","国产 AI 桌面客户端第一梯队，多模型聚合 + 本地 RAG + 中文体验顶级。需要 Web 部署 \u002F 自托管选 LobeChat；只要桌面体验完整选 Cherry Studio。","https:\u002F\u002Fcherry-ai.com","CVmnny2iirvdFz56djfCs7oCOn1Mjv6hZCZ5SNO0U0Q",{"id":984,"title":419,"alternatives":985,"api_compatible":15,"body":986,"category":470,"chinese_friendly":459,"cover":1370,"description":1371,"domestic":473,"extension":474,"faq":15,"free":473,"github":1355,"languages":1372,"lastVerified":477,"meta":1373,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":1374,"pillar":481,"platforms":1375,"priceTable":15,"pricing":1376,"published":487,"relatedPlaybooks":15,"relatedReviews":15,"score":1377,"self_host":473,"seo":1378,"seoTitle":1379,"slug":1380,"sources":1381,"stem":1384,"suitable":15,"tagline":1385,"tags":1386,"updated":477,"verdict":1388,"website":1349,"__hash__":1389},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fgpt4all.md",[14,12,493],{"type":17,"value":987,"toc":1357},[988,990,997,1000,1002,1051,1053,1057,1080,1083,1085,1091,1108,1112,1129,1131,1155,1157,1248,1250,1282,1284,1306,1308,1314,1320,1326,1328,1337,1339,1343],[20,989,23],{"id":22},[25,991,992,993,996],{},"GPT4All 是 Nomic AI 出品的本地 LLM 推理工具，最大卖点是",[29,994,995],{},"CPU 也能跑","——没有独立显卡的普通笔记本 \u002F 办公本照样运行本地大模型。MIT 开源，桌面客户端一键下载 GGUF 模型，开箱即用。适合无 GPU 设备、隐私优先、轻量本地推理场景。",[25,998,999],{},"适合：没有独立 GPU 的笔记本 \u002F 办公本用户、需要完全离线本地推理、隐私敏感场景、轻量聊天 \u002F 文档处理。不适合：需要高速推理（CPU 太慢）、需要 OpenAI 兼容 API 给应用接入（用 Ollama）、需要大模型（32B+，CPU 跑不动）。",[20,1001,38],{"id":38},[40,1003,1004,1010,1015,1021,1027,1033,1039,1045],{},[43,1005,1006,1009],{},[29,1007,1008],{},"CPU 推理","：基于 llama.cpp 优化，纯 CPU 跑 7B 量化模型，无需 GPU",[43,1011,1012,1014],{},[29,1013,53],{},"：内置模型库，点击即下载 GGUF 格式模型，自动配置",[43,1016,1017,1020],{},[29,1018,1019],{},"桌面客户端","：跨平台 GUI（Win \u002F Mac \u002F Linux），聊天界面开箱即用",[43,1022,1023,1026],{},[29,1024,1025],{},"LocalDocs（RAG）","：内置文档问答功能，拖入 PDF \u002F 文档即可基于本地文档聊天",[43,1028,1029,1032],{},[29,1030,1031],{},"OpenAI 兼容 API","：内置 Local Server，暴露 OpenAI 兼容端点供应用调用",[43,1034,1035,1038],{},[29,1036,1037],{},"模型库丰富","：Llama \u002F Qwen \u002F Mistral \u002F Phi \u002F GPT-OSS 等主流开源模型可选",[43,1040,1041,1044],{},[29,1042,1043],{},"GPU 加速（可选）","：有 GPU 时自动启用，速度提升数倍",[43,1046,1047,1050],{},[29,1048,1049],{},"MIT 开源","：完全免费，可商用，提供 Python \u002F C++ SDK 二次开发",[20,1052,93],{"id":93},[95,1054,1055],{},[25,1056,99],{},[101,1058,1059,1069],{},[104,1060,1061],{},[107,1062,1063,1065,1067],{},[110,1064,112],{},[110,1066,93],{},[110,1068,117],{},[119,1070,1071],{},[107,1072,1073,1075,1077],{},[124,1074,126],{},[124,1076,129],{},[124,1078,1079],{},"完整功能，MIT 协议，商用免费",[25,1081,1082],{},"完全免费。成本在于硬件（CPU \u002F 内存）和你选用的模型。",[20,1084,139],{"id":138},[95,1086,1087],{},[25,1088,144,1089],{},[29,1090,147],{},[40,1092,1093,1096,1099,1102,1105],{},[43,1094,1095],{},"纯 CPU 跑 Qwen2.5-7B-Q4 在 i7 笔记本上约 5-8 tok\u002Fs，轻量聊天可用",[43,1097,1098],{},"LocalDocs 文档问答好用：拖入技术 PDF，直接问问题，完全离线",[43,1100,1101],{},"模型一键下载体验顺滑，不用手动找 GGUF + 配路径",[43,1103,1104],{},"有 GPU 时自动加速，RTX 3060 跑 7B 约 30-40 tok\u002Fs",[43,1106,1107],{},"桌面 GUI 简洁易用，非技术用户也能上手",[25,1109,1110],{},[29,1111,169],{},[40,1113,1114,1117,1120,1123,1126],{},[43,1115,1116],{},"CPU 推理速度慢，7B 模型生成一段代码要等 10-20 秒",[43,1118,1119],{},"内存占用高：7B-Q4 至少需 8GB RAM，13B 需 16GB",[43,1121,1122],{},"LocalDocs 的 RAG 质量一般，复杂文档检索准确率不高",[43,1124,1125],{},"OpenAI 兼容 API 功能弱，不如 Ollama 灵活，不支持自定义 Modelfile",[43,1127,1128],{},"中文模型支持一般，需手动选 Qwen 等中文友好的模型",[20,1130,189],{"id":189},[191,1132,1133,1136,1139,1142,1145,1148],{},[43,1134,1135],{},"从 gpt4all.io 下载对应平台安装包",[43,1137,1138],{},"安装后打开桌面客户端",[43,1140,1141],{},"点击 \"Downloads\" → 选模型（推荐 Qwen2.5-7B-Q4 或 Llama3.1-8B-Q4）",[43,1143,1144],{},"下载完成后回到 Chat → 选模型 → 开始聊天",[43,1146,1147],{},"文档问答：LocalDocs → 添加文档文件夹 → 在聊天中勾选引用",[43,1149,1150,1151],{},"API 接入：Settings → 启用 API Server → 端点 ",[1152,1153,1154],"code",{},"http:\u002F\u002Flocalhost:4891\u002Fv1",[20,1156,213],{"id":213},[101,1158,1159,1171],{},[104,1160,1161],{},[107,1162,1163,1165,1167,1169],{},[110,1164,222],{},[110,1166,419],{},[110,1168,230],{},[110,1170,227],{},[119,1172,1173,1184,1197,1211,1223,1237],{},[107,1174,1175,1177,1180,1182],{},[124,1176,1008],{},[124,1178,1179],{},"✅ 优化好",[124,1181,276],{},[124,1183,276],{},[107,1185,1186,1189,1192,1195],{},[124,1187,1188],{},"GUI",[124,1190,1191],{},"✅ 桌面",[124,1193,1194],{},"❌ CLI",[124,1196,276],{},[107,1198,1199,1202,1205,1208],{},[124,1200,1201],{},"模型下载",[124,1203,1204],{},"✅ 内置库",[124,1206,1207],{},"✅ CLI",[124,1209,1210],{},"✅ GUI 库",[107,1212,1213,1216,1218,1221],{},[124,1214,1215],{},"OpenAI API",[124,1217,273],{},[124,1219,1220],{},"强（Modelfile）",[124,1222,276],{},[107,1224,1225,1228,1231,1234],{},[124,1226,1227],{},"文档 RAG",[124,1229,1230],{},"✅ LocalDocs",[124,1232,1233],{},"需配 WebUI",[124,1235,1236],{},"需插件",[107,1238,1239,1242,1244,1246],{},[124,1240,1241],{},"开源",[124,1243,322],{},[124,1245,322],{},[124,1247,319],{},[20,1249,328],{"id":328},[40,1251,1252,1258,1264,1270,1276],{},[43,1253,1254,1257],{},[29,1255,1256],{},"CPU 用户别跑大模型","：13B 以上 CPU 跑基本不可用，坚持 7B 量化",[43,1259,1260,1263],{},[29,1261,1262],{},"LocalDocs 别放太多文件","：文件多了索引慢且检索质量下降，分批放",[43,1265,1266,1269],{},[29,1267,1268],{},"API Server 别对公网开","：默认无鉴权，仅本地用",[43,1271,1272,1275],{},[29,1273,1274],{},"内存不够会崩","：8GB RAM 只能跑 7B-Q4，别勉强 13B",[43,1277,1278,1281],{},[29,1279,1280],{},"中文场景选 Qwen","：Llama 系列中文能力弱，Qwen \u002F GLM 更合适",[20,1283,364],{"id":363},[40,1285,1286,1289,1292,1295,1298,1301,1303],{},[43,1287,1288],{},"✅ 没有独立 GPU 的笔记本 \u002F 办公本",[43,1290,1291],{},"✅ 需要完全离线 + 隐私优先的本地推理",[43,1293,1294],{},"✅ 轻量聊天 \u002F 文档问答场景",[43,1296,1297],{},"✅ 非技术用户（GUI 友好）",[43,1299,1300],{},"❌ 需要高速推理（有 GPU 直接 Ollama）",[43,1302,381],{},[43,1304,1305],{},"❌ 需要跑大模型（32B+，CPU 跑不动）",[20,1307,391],{"id":390},[25,1309,1310,1313],{},[29,1311,1312],{},"Q: CPU 跑 7B 模型速度能接受吗？","\nA: 看用途。轻量聊天 \u002F 短问答可以接受（5-8 tok\u002Fs）。但代码生成 \u002F 长文本输出等待时间较长（一段代码 10-20 秒）。有 GPU 强烈建议用 GPU。",[25,1315,1316,1319],{},[29,1317,1318],{},"Q: GPT4All 和 Ollama 怎么选？","\nA: 没 GPU + 要 GUI → GPT4All。有 GPU + 要 API 接入应用 → Ollama。两者底层都基于 llama.cpp，模型格式通用（GGUF）。",[25,1321,1322,1325],{},[29,1323,1324],{},"Q: LocalDocs 文档问答好用吗？","\nA: 基础可用但 RAG 质量一般。简单 PDF 问答没问题，复杂文档 \u002F 多文件检索准确率不高。专业 RAG 需求建议配 Open WebUI + 向量数据库。",[20,1327,412],{"id":412},[25,1329,1330,420,1333,420,1335],{},[416,1331,10],{"href":1332},"\u002Fcoding\u002Flocal\u002Fjan.html",[416,1334,424],{"href":423},[416,1336,428],{"href":427},[20,1338,431],{"id":431},[95,1340,1341],{},[25,1342,436],{},[40,1344,1345,1351],{},[43,1346,1347],{},[416,1348,446],{"href":1349,"rel":1350},"https:\u002F\u002Fgpt4all.io",[445],[43,1352,1353],{},[416,1354,453],{"href":1355,"rel":1356},"https:\u002F\u002Fgithub.com\u002Fnomic-ai\u002Fgpt4all",[445],{"title":455,"searchDepth":456,"depth":456,"links":1358},[1359,1360,1361,1362,1363,1364,1365,1366,1367,1368,1369],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":138,"depth":459,"text":139},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":390,"depth":459,"text":391},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Fgpt4all.webp","GPT4All 真实评测：Nomic AI 出品的本地 LLM 推理工具（MIT 协议），支持 CPU 运行无需 GPU，GGUF 格式模型一键下载。跨平台桌面客户端，适合没有独立显卡、需要在普通笔记本上跑本地大模型的用户。",[476],{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fgpt4all",[483,484,485],"Free \u002F 开源（MIT）",{"power":456,"ux":489,"price":490,"cn_support":459,"stability":489},{"title":419,"description":1371},"GPT4All - 本地 LLM 推理引擎评测与使用 | AIHO","coding\u002Flocal\u002Fgpt4all",[1382,1383],{"title":446,"url":1349},{"title":453,"url":1355},"tools\u002Fcoding\u002Flocal\u002Fgpt4all","本地 LLM 推理，Nomic AI 出品，支持 CPU 运行",[470,1387,501,502],"cpu-inference","没有 GPU 也能跑本地大模型的首选，CPU 推理 + 一键下载模型体验顺滑；但推理速度慢、API 能力弱，有 GPU 用户建议直接 Ollama。","oTtyXhNkQBIi7aJI5YJrsiu_c1A81ifErtlA5QOs9UI",{"id":9,"title":10,"alternatives":1391,"api_compatible":15,"body":1392,"category":470,"chinese_friendly":456,"cover":471,"description":472,"domestic":473,"extension":474,"faq":15,"free":473,"github":451,"languages":1709,"lastVerified":477,"meta":1710,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":480,"pillar":481,"platforms":1711,"priceTable":15,"pricing":486,"published":487,"relatedPlaybooks":15,"relatedReviews":15,"score":1712,"self_host":473,"seo":1713,"seoTitle":492,"slug":493,"sources":1714,"stem":497,"suitable":15,"tagline":498,"tags":1717,"updated":477,"verdict":504,"website":443,"__hash__":505},[12,13,14],{"type":17,"value":1393,"toc":1696},[1394,1396,1400,1402,1404,1438,1440,1444,1466,1468,1470,1476,1488,1492,1504,1506,1520,1522,1612,1614,1636,1638,1654,1656,1660,1664,1668,1670,1678,1680,1684],[20,1395,23],{"id":22},[25,1397,27,1398,32],{},[29,1399,31],{},[25,1401,35],{},[20,1403,38],{"id":38},[40,1405,1406,1410,1414,1418,1422,1426,1430,1434],{},[43,1407,1408,48],{},[29,1409,47],{},[43,1411,1412,54],{},[29,1413,53],{},[43,1415,1416,60],{},[29,1417,59],{},[43,1419,1420,66],{},[29,1421,65],{},[43,1423,1424,72],{},[29,1425,71],{},[43,1427,1428,78],{},[29,1429,77],{},[43,1431,1432,84],{},[29,1433,83],{},[43,1435,1436,90],{},[29,1437,89],{},[20,1439,93],{"id":93},[95,1441,1442],{},[25,1443,99],{},[101,1445,1446,1456],{},[104,1447,1448],{},[107,1449,1450,1452,1454],{},[110,1451,112],{},[110,1453,93],{},[110,1455,117],{},[119,1457,1458],{},[107,1459,1460,1462,1464],{},[124,1461,126],{},[124,1463,129],{},[124,1465,132],{},[25,1467,135],{},[20,1469,139],{"id":138},[95,1471,1472],{},[25,1473,144,1474],{},[29,1475,147],{},[40,1477,1478,1480,1482,1484,1486],{},[43,1479,152],{},[43,1481,155],{},[43,1483,158],{},[43,1485,161],{},[43,1487,164],{},[25,1489,1490],{},[29,1491,169],{},[40,1493,1494,1496,1498,1500,1502],{},[43,1495,174],{},[43,1497,177],{},[43,1499,180],{},[43,1501,183],{},[43,1503,186],{},[20,1505,189],{"id":189},[191,1507,1508,1510,1512,1514,1516,1518],{},[43,1509,195],{},[43,1511,198],{},[43,1513,201],{},[43,1515,204],{},[43,1517,207],{},[43,1519,210],{},[20,1521,213],{"id":213},[101,1523,1524,1538],{},[104,1525,1526],{},[107,1527,1528,1530,1532,1534,1536],{},[110,1529,222],{},[110,1531,10],{},[110,1533,227],{},[110,1535,230],{},[110,1537,233],{},[119,1539,1540,1552,1564,1576,1588,1600],{},[107,1541,1542,1544,1546,1548,1550],{},[124,1543,240],{},[124,1545,243],{},[124,1547,246],{},[124,1549,249],{},[124,1551,243],{},[107,1553,1554,1556,1558,1560,1562],{},[124,1555,256],{},[124,1557,246],{},[124,1559,261],{},[124,1561,261],{},[124,1563,246],{},[107,1565,1566,1568,1570,1572,1574],{},[124,1567,270],{},[124,1569,273],{},[124,1571,276],{},[124,1573,261],{},[124,1575,276],{},[107,1577,1578,1580,1582,1584,1586],{},[124,1579,285],{},[124,1581,276],{},[124,1583,290],{},[124,1585,290],{},[124,1587,276],{},[107,1589,1590,1592,1594,1596,1598],{},[124,1591,299],{},[124,1593,276],{},[124,1595,276],{},[124,1597,306],{},[124,1599,276],{},[107,1601,1602,1604,1606,1608,1610],{},[124,1603,313],{},[124,1605,316],{},[124,1607,319],{},[124,1609,322],{},[124,1611,325],{},[20,1613,328],{"id":328},[40,1615,1616,1620,1624,1628,1632],{},[43,1617,1618,336],{},[29,1619,335],{},[43,1621,1622,342],{},[29,1623,341],{},[43,1625,1626,348],{},[29,1627,347],{},[43,1629,1630,354],{},[29,1631,353],{},[43,1633,1634,360],{},[29,1635,359],{},[20,1637,364],{"id":363},[40,1639,1640,1642,1644,1646,1648,1650,1652],{},[43,1641,369],{},[43,1643,372],{},[43,1645,375],{},[43,1647,378],{},[43,1649,381],{},[43,1651,384],{},[43,1653,387],{},[20,1655,391],{"id":390},[25,1657,1658,397],{},[29,1659,396],{},[25,1661,1662,403],{},[29,1663,402],{},[25,1665,1666,409],{},[29,1667,408],{},[20,1669,412],{"id":412},[25,1671,1672,420,1674,420,1676],{},[416,1673,419],{"href":418},[416,1675,424],{"href":423},[416,1677,428],{"href":427},[20,1679,431],{"id":431},[95,1681,1682],{},[25,1683,436],{},[40,1685,1686,1691],{},[43,1687,1688],{},[416,1689,446],{"href":443,"rel":1690},[445],[43,1692,1693],{},[416,1694,453],{"href":451,"rel":1695},[445],{"title":455,"searchDepth":456,"depth":456,"links":1697},[1698,1699,1700,1701,1702,1703,1704,1705,1706,1707,1708],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":138,"depth":459,"text":139},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":390,"depth":459,"text":391},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},[476],{},[483,484,485],{"power":456,"ux":489,"price":490,"cn_support":456,"stability":456},{"title":10,"description":472},[1715,1716],{"title":446,"url":443},{"title":453,"url":451},[470,500,501,502,503],{"id":1719,"title":227,"alternatives":1720,"api_compatible":15,"body":1721,"category":470,"chinese_friendly":456,"cover":2145,"description":2146,"domestic":473,"extension":474,"faq":2147,"free":473,"github":15,"languages":2160,"lastVerified":15,"meta":2161,"models":15,"navigation":479,"notSuitable":15,"opensource":473,"path":875,"pillar":481,"platforms":2162,"priceTable":2163,"pricing":2171,"published":957,"relatedPlaybooks":2172,"relatedReviews":15,"score":2174,"self_host":479,"seo":2175,"seoTitle":2176,"slug":12,"sources":2177,"stem":2184,"suitable":15,"tagline":2185,"tags":2186,"updated":967,"verdict":2192,"website":2193,"__hash__":2194},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Flm-studio.md",[14,511,13,510],{"type":17,"value":1722,"toc":2133},[1723,1725,1732,1735,1737,1794,1796,1810,1815,1819,1823,1840,1844,1861,1863,1889,1891,2024,2026,2058,2060,2083,2085,2108,2110],[20,1724,23],{"id":22},[25,1726,1727,1728,1731],{},"LM Studio 是 Windows \u002F macOS \u002F Linux 桌面应用，让你像浏览 App Store 一样发现、下载、运行本地大模型（GGUF \u002F MLX 格式）。底层基于 llama.cpp + MLX，Mac M 系列原生优化。0.3+ 起新增 Headless 模式 + ",[1152,1729,1730],{},"lms"," CLI，可在服务器跑 OpenAI 兼容 API（默认 :1234）。个人 \u002F 评估完全免费，商用咨询。",[25,1733,1734],{},"适合：本地 LLM 入门 \u002F 评估、Mac 用户、需要 GUI 调参 \u002F 模型比较、想给 IDE \u002F 应用接本地 OpenAI 兼容 endpoint 的开发者。不适合：多用户并发生产服务（用 vLLM）、嵌入式 \u002F 边缘部署（用 llama.cpp）、纯 CLI 工作流（用 Ollama）。",[20,1736,38],{"id":38},[40,1738,1739,1745,1751,1757,1767,1776,1782,1788],{},[43,1740,1741,1744],{},[29,1742,1743],{},"模型浏览器","：内置 Hugging Face 检索，按 GGUF \u002F MLX \u002F 大小筛选、一键下载",[43,1746,1747,1750],{},[29,1748,1749],{},"聊天界面","：System Prompt \u002F temperature \u002F top-p \u002F context size 可视化调参",[43,1752,1753,1756],{},[29,1754,1755],{},"多模型并存 \u002F 切换","：同时加载多模型在不同会话中比较",[43,1758,1759,1762,1763,1766],{},[29,1760,1761],{},"OpenAI 兼容 Local Server","：",[1152,1764,1765],{},"http:\u002F\u002Flocalhost:1234\u002Fv1","，任何 SDK 即接即用",[43,1768,1769,1762,1772,1775],{},[29,1770,1771],{},"Headless \u002F CLI",[1152,1773,1774],{},"lms server start --port 1234","，无 GUI 可跑",[43,1777,1778,1781],{},[29,1779,1780],{},"PDF \u002F 文档对话","：内置基础 RAG，丢文件就能聊",[43,1783,1784,1787],{},[29,1785,1786],{},"MLX 原生支持（Mac）","：M1+ 上比 GGUF + Metal 快 30–50%",[43,1789,1790,1793],{},[29,1791,1792],{},"持续批处理","：Codersera 2026 测得 50–90 tok\u002Fs（消费级 GPU + 中等模型）",[20,1795,93],{"id":93},[40,1797,1798,1804],{},[43,1799,1800,1803],{},[29,1801,1802],{},"个人 \u002F 评估","：免费，全功能可用",[43,1805,1806,1809],{},[29,1807,1808],{},"商用","：邮件 \u002F 官网联系 LM Studio 团队",[95,1811,1812],{},[25,1813,1814],{},"模型本身免费（开源权重），LM Studio 不抽水任何 token 费用。",[20,1816,1818],{"id":1817},"实测mac-m2-pro-qwen3-coder-7b-gguf-q4_k_m","实测（Mac M2 Pro + Qwen3-Coder-7B GGUF Q4_K_M）",[25,1820,1821],{},[29,1822,147],{},[40,1824,1825,1828,1831,1834,1837],{},[43,1826,1827],{},"模型浏览器极舒服：搜「qwen3-coder」直接列出 GGUF + MLX 各 quant，标硬件兼容度",[43,1829,1830],{},"加载 7B Q4 模型 \u003C 3 秒，生成 ~75 tok\u002Fs",[43,1832,1833],{},"Local Server 开了 Cursor 直接接 baseURL → 本地代码补全零成本",[43,1835,1836],{},"MLX 版同模型 ~110 tok\u002Fs，差距显著",[43,1838,1839],{},"多窗口加载 2 个模型并排测，调 prompt 直观",[25,1841,1842],{},[29,1843,169],{},[40,1845,1846,1849,1852,1855,1858],{},[43,1847,1848],{},"模型库依赖 Hugging Face，国内访问要镜像 \u002F 代理",[43,1850,1851],{},"GPU 显存吃满后会自动 offload 到 CPU，无提示就慢下来",[43,1853,1854],{},"Headless 模式相对 Ollama 偏新，文档稍少",[43,1856,1857],{},"闭源应用（虽免费），不适合企业合规挂钩",[43,1859,1860],{},"中文 UI 可用但部分菜单仍英文",[20,1862,189],{"id":189},[191,1864,1865,1868,1871,1874,1877,1884],{},[43,1866,1867],{},"lmstudio.ai 下载（Mac \u002F Windows \u002F Linux）",[43,1869,1870],{},"打开 → Discover 标签 → 搜模型（如 qwen3-coder、deepseek-v3 GGUF\u002FMLX）→ Download",[43,1872,1873],{},"Chat 标签 → 选模型 → 调参聊天",[43,1875,1876],{},"Local Server 标签 → Start Server → 默认端口 1234",[43,1878,1879,1880,1883],{},"在你的应用里：",[1152,1881,1882],{},"baseURL = \"http:\u002F\u002Flocalhost:1234\u002Fv1\"","，API Key 任意",[43,1885,1886,1887],{},"Headless：",[1152,1888,1774],{},[20,1890,213],{"id":213},[101,1892,1893,1908],{},[104,1894,1895],{},[107,1896,1897,1899,1901,1903,1905],{},[110,1898,222],{},[110,1900,227],{},[110,1902,230],{},[110,1904,679],{},[110,1906,1907],{},"llama.cpp",[119,1909,1910,1926,1942,1956,1970,1984,1997,2009],{},[107,1911,1912,1914,1917,1920,1923],{},[124,1913,686],{},[124,1915,1916],{},"GUI + CLI",[124,1918,1919],{},"CLI Daemon",[124,1921,1922],{},"Docker UI",[124,1924,1925],{},"二进制",[107,1927,1928,1931,1934,1937,1939],{},[124,1929,1930],{},"模型浏览",[124,1932,1933],{},"✅ 内置",[124,1935,1936],{},"CLI pull",[124,1938,753],{},[124,1940,1941],{},"手动",[107,1943,1944,1947,1949,1951,1954],{},[124,1945,1946],{},"参数调优 GUI",[124,1948,276],{},[124,1950,290],{},[124,1952,1953],{},"部分",[124,1955,290],{},[107,1957,1958,1960,1963,1966,1968],{},[124,1959,1031],{},[124,1961,1962],{},"✅ :1234",[124,1964,1965],{},"✅ :11434",[124,1967,276],{},[124,1969,276],{},[107,1971,1972,1975,1977,1980,1982],{},[124,1973,1974],{},"MLX (Mac)",[124,1976,276],{},[124,1978,1979],{},"✅ 0.19+",[124,1981,798],{},[124,1983,798],{},[107,1985,1986,1989,1991,1993,1995],{},[124,1987,1988],{},"多用户并发",[124,1990,724],{},[124,1992,724],{},[124,1994,276],{},[124,1996,246],{},[107,1998,1999,2001,2003,2005,2007],{},[124,2000,1241],{},[124,2002,782],{},[124,2004,322],{},[124,2006,322],{},[124,2008,322],{},[107,2010,2011,2014,2017,2020,2022],{},[124,2012,2013],{},"上手难度",[124,2015,2016],{},"极低",[124,2018,2019],{},"低",[124,2021,246],{},[124,2023,243],{},[20,2025,328],{"id":328},[40,2027,2028,2034,2040,2046,2052],{},[43,2029,2030,2033],{},[29,2031,2032],{},"国内下模型走镜像","：HF 直连慢 \u002F 卡，配 HF_ENDPOINT=hf-mirror.com",[43,2035,2036,2039],{},[29,2037,2038],{},"显存爆 ≠ 报错","：GPU 装不下会无声 offload 到 CPU，关注生成速度，必要时降 quant 或换小模型",[43,2041,2042,2045],{},[29,2043,2044],{},"MLX 优先（Mac M 系列）","：能下 MLX 版就别下 GGUF，速度差距明显",[43,2047,2048,2051],{},[29,2049,2050],{},"Local Server 暴露要谨慎","：默认 0.0.0.0 + 无鉴权，对外开放前加反代 + Bearer",[43,2053,2054,2057],{},[29,2055,2056],{},"闭源合规要核","：企业内部使用前查 license；商用必须联系官方",[20,2059,364],{"id":363},[40,2061,2062,2065,2068,2071,2074,2077,2080],{},[43,2063,2064],{},"✅ 本地 LLM 入门 \u002F 评估",[43,2066,2067],{},"✅ Mac M 系列用户",[43,2069,2070],{},"✅ 想给 Cursor \u002F Cline 接本地 OpenAI 兼容 endpoint",[43,2072,2073],{},"✅ 需要 GUI 调参 \u002F 模型比较",[43,2075,2076],{},"❌ 多用户并发生产服务",[43,2078,2079],{},"❌ 嵌入式 \u002F 边缘设备",[43,2081,2082],{},"❌ 强合规 \u002F 必须开源审计",[20,2084,412],{"id":412},[40,2086,2087,2091,2097,2102],{},[43,2088,2089],{},[416,2090,882],{"href":881},[43,2092,2093],{},[416,2094,2096],{"href":2095},"\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui","Open WebUI 评测",[43,2098,2099],{},[416,2100,2101],{"href":944},"Cherry Studio 评测",[43,2103,2104],{},[416,2105,2107],{"href":2106},"\u002Fplaybook\u002Fonboarding\u002Fclaude-code-getting-started","Claude Code 上手 Playbook",[20,2109,431],{"id":431},[191,2111,2112,2119,2126],{},[43,2113,2114,2115],{},"LM Studio 官网 ",[416,2116,2117],{"href":2117,"rel":2118},"https:\u002F\u002Flmstudio.ai\u002F",[445],[43,2120,2121,2122],{},"Codersera — LM Studio Complete Guide 2026 ",[416,2123,2124],{"href":2124,"rel":2125},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Flm-studio-complete-guide-2026\u002F",[445],[43,2127,2128,2129],{},"Codersera — Ollama vs LM Studio vs vLLM vs llama.cpp vs MLX 2026 ",[416,2130,2131],{"href":2131,"rel":2132},"https:\u002F\u002Fcodersera.com\u002Fblog\u002Follama-vs-lm-studio-vs-vllm-vs-llama-cpp-vs-mlx-2026\u002F",[445],{"title":455,"searchDepth":456,"depth":456,"links":2134},[2135,2136,2137,2138,2139,2140,2141,2142,2143,2144],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":1817,"depth":459,"text":1818},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Flm-studio.webp","LM Studio 真实评测：跨平台桌面应用，运行本地 GGUF \u002F MLX 大模型。50–90 tok\u002Fs 持续批处理、OpenAI 兼容本地 API（默认端口 1234）、Headless 模式、Mac \u002F Win 双端。对个人开发者免费，企业咨询。",[2148,2151,2154,2157],{"q":2149,"a":2150},"和 Ollama 怎么选？","LM Studio 是 GUI 优先（模型浏览器 + 参数面板 + 聊天界面），适合个人 \u002F 评估 \u002F 上手。Ollama 是 CLI \u002F Daemon 优先（后台跑 + REST API），适合应用嵌入 \u002F 脚本调用。两者都基于 llama.cpp，在 Mac M 系列上都已用 MLX。",{"q":2152,"a":2153},"支持 MLX 吗？","支持。Mac M1+ 上可加载 MLX 格式模型，速度比 GGUF + Metal 快 30–50%。模型搜索时筛选 MLX 即可。",{"q":2155,"a":2156},"OpenAI 兼容 API 怎么用？","开 Local Server → 默认端口 1234 → `http:\u002F\u002Flocalhost:1234\u002Fv1`。任何 OpenAI SDK 把 baseURL 改这个就能跑本地模型，零代码改动。",{"q":2158,"a":2159},"Headless 模式？","0.3+ 起支持 `lms server start` CLI 启动后台服务，无 GUI 即可跑 OpenAI 兼容 API，适合服务器 \u002F SSH 场景。",[476,942],{},[483,484,485],[2164,2167],{"plan":1802,"price":949,"features":2165,"notes":2166},"全功能 GUI + Headless API + GGUF\u002FMLX","供个人 \u002F 评估使用",{"plan":1808,"price":2168,"features":2169,"notes":2170},"联系咨询","团队部署 \u002F 商用 license","邮件 \u002F 官网联系","免费（个人 \u002F 评估） \u002F 企业 \u002F 商用咨询",[959,2173],"onboarding\u002Fclaude-code-getting-started",{"power":489,"ux":490,"price":490,"cn_support":456,"stability":489},{"title":227,"description":2146},"LM Studio 评测 2026：本地运行开源大模型，图形化界面，AI 模型管理",[2178,2180,2182],{"name":2179,"url":2117,"accessed":967},"LM Studio 官网",{"name":2181,"url":2124,"accessed":967},"Codersera — LM Studio Complete Guide 2026",{"name":2183,"url":2131,"accessed":967},"Codersera — Ollama vs LM Studio vs vLLM 2026","tools\u002Fcoding\u002Flocal\u002Flm-studio","本地 LLM 的 GUI 首选——模型浏览器 + GGUF\u002FMLX 推理 + OpenAI 兼容 API + Mac 原生优化",[470,2187,502,2188,2189,2190,2191],"gui","mlx","llama-cpp","mac","openai-compatible","Mac \u002F Windows 桌面本地 LLM 的 GUI 首选——上手最快、模型浏览最舒服、自带 OpenAI 兼容 API。批量服务 \u002F 多用户场景用 vLLM；纯 CLI \u002F 嵌入应用走 Ollama。","https:\u002F\u002Flmstudio.ai","LobnLABcHoL2A6Bu-tfBMAlwTLq_jKyEJkNmD662RWU",{"id":2196,"title":674,"alternatives":2197,"api_compatible":15,"body":2198,"category":470,"chinese_friendly":490,"cover":2634,"description":2635,"domestic":473,"extension":474,"faq":2636,"free":473,"github":15,"languages":2649,"lastVerified":15,"meta":2650,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":869,"pillar":481,"platforms":2651,"priceTable":2654,"pricing":2662,"published":957,"relatedPlaybooks":2663,"relatedReviews":15,"score":2664,"self_host":479,"seo":2665,"seoTitle":2666,"slug":510,"sources":2667,"stem":2674,"suitable":15,"tagline":2675,"tags":2676,"updated":967,"verdict":2679,"website":2680,"__hash__":2681},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Flobe-chat.md",[13,511,14,12],{"type":17,"value":2199,"toc":2622},[2200,2202,2209,2212,2214,2276,2278,2291,2294,2298,2302,2322,2326,2343,2345,2371,2373,2509,2511,2549,2551,2577,2579,2597,2599],[20,2201,23],{"id":22},[25,2203,2204,2205,2208],{},"LobeChat 是 LobeHub 团队的开源 AI 聊天框架，2023 年发布、GitHub 72k+ stars、MIT 协议。",[29,2206,2207],{},"Web + 桌面 + Docker 自托管三形态","，把 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F Ollama \u002F LM Studio 等 80+ 模型聚合到一个现代设计的客户端里。内置 RAG 知识库 + 插件市场 + 助手市场 + 多模型对比 + MCP，是当下综合最强的多模型 AI 客户端之一。",[25,2210,2211],{},"适合：需要 Web 端访问、Docker 自托管、多模型对比、丰富助手市场的用户；中文重度用户；想给团队 \u002F 家庭部署一个共享 AI 工作台。不适合：只用桌面 + 不需要 Web（Cherry Studio 同样优秀且更精细）、强企业 RBAC + 多租户（Open WebUI 多用户更完善）。",[20,2213,38],{"id":38},[40,2215,2216,2221,2227,2232,2238,2244,2249,2254,2260,2266],{},[43,2217,2218,2220],{},[29,2219,530],{},"：OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Kimi \u002F 豆包 \u002F Groq \u002F Together \u002F OpenRouter \u002F Ollama \u002F LM Studio",[43,2222,2223,2226],{},[29,2224,2225],{},"多模型对比","：同 prompt 给多模型并排回答",[43,2228,2229,2231],{},[29,2230,536],{},"：上传 PDF \u002F Word \u002F 网页 → 向量化 → 检索引用",[43,2233,2234,2237],{},[29,2235,2236],{},"插件市场","：联网搜索 \u002F 代码执行 \u002F 图像生成 \u002F 翻译等几十款官方插件",[43,2239,2240,2243],{},[29,2241,2242],{},"助手市场","：几百个预设 AI 角色，一键导入",[43,2245,2246,2248],{},[29,2247,548],{},"：扩展任意工具能力",[43,2250,2251],{},[29,2252,2253],{},"代码解释器 \u002F 文件上传 \u002F TTS \u002F 多模态",[43,2255,2256,2259],{},[29,2257,2258],{},"Web + 桌面 + Docker","：三形态，数据可完全本地",[43,2261,2262,2265],{},[29,2263,2264],{},"LobeHub Cloud","：官方云托管，免部署",[43,2267,2268,2271,2272,2275],{},[29,2269,2270],{},"快捷指令 \u002F 工作流","：自定义 prompt 模板，",[1152,2273,2274],{},"\u002Fpodcast-summary"," 类用法",[20,2277,93],{"id":93},[40,2279,2280,2286],{},[43,2281,2282,2285],{},[29,2283,2284],{},"自托管 \u002F 桌面","：完全免费、MIT 开源",[43,2287,2288,2290],{},[29,2289,2264],{},"：订阅制，云端托管 + 团队协作 + 同步",[25,2292,2293],{},"模型 API 费用按你自己的供应商付费；本地 Ollama \u002F LM Studio 零成本。",[20,2295,2297],{"id":2296},"实测m2-自托管-docker连-openai-deepseek-本地-ollama","实测（M2 + 自托管 Docker，连 OpenAI + DeepSeek + 本地 Ollama）",[25,2299,2300],{},[29,2301,147],{},[40,2303,2304,2307,2310,2313,2316,2319],{},[43,2305,2306],{},"界面颜值是这一类工具里第一档（深色 \u002F 透明 \u002F 现代感）",[43,2308,2309],{},"多模型并排对比对选型极其有用：写一道复杂题，Claude \u002F GPT \u002F DeepSeek 直接对比答案",[43,2311,2312],{},"知识库 RAG 上传 50+ PDF 后检索准确，引用片段可视化",[43,2314,2315],{},"助手市场拿来即用——「Code Reviewer」「Translation Polish」节省 prompt 编写",[43,2317,2318],{},"Docker 一键部署，团队 5 人共享流畅",[43,2320,2321],{},"多平台数据同步（Cloud \u002F WebDAV）",[25,2323,2324],{},[29,2325,169],{},[40,2327,2328,2331,2334,2337,2340],{},[43,2329,2330],{},"自托管要熟悉 Docker + 反代 + HTTPS",[43,2332,2333],{},"国内连 OpenAI \u002F Claude 需自带网络方案",[43,2335,2336],{},"Web 版数据存 LobeHub，隐私敏感场景走桌面 \u002F Docker",[43,2338,2339],{},"插件市场质量参差，要自筛",[43,2341,2342],{},"团队多人共享需配 LobeHub Cloud 或自建数据库（Postgres + S3）",[20,2344,189],{"id":189},[191,2346,2347,2350,2356,2359,2362,2365,2368],{},[43,2348,2349],{},"选形态：Web（chat.lobehub.com 注册即用） \u002F 桌面（GitHub Releases 下载） \u002F Docker",[43,2351,2352,2353],{},"Docker：",[1152,2354,2355],{},"docker run -d -p 3210:3210 -e OPENAI_API_KEY=sk-xxx --name lobe-chat lobehub\u002Flobe-chat",[43,2357,2358],{},"设置 → AI 服务商 → 添加 OpenAI \u002F Claude \u002F DeepSeek \u002F Ollama",[43,2360,2361],{},"模型选择器测试对话",[43,2363,2364],{},"知识库：拖文件 → 等向量化 → 对话引用",[43,2366,2367],{},"助手市场拉「Code Reviewer」「论文翻译润色」试用",[43,2369,2370],{},"进阶：插件市场启用联网搜索 \u002F 代码执行；MCP 自定义工具",[20,2372,213],{"id":213},[101,2374,2375,2389],{},[104,2376,2377],{},[107,2378,2379,2381,2383,2385,2387],{},[110,2380,222],{},[110,2382,674],{},[110,2384,233],{},[110,2386,679],{},[110,2388,227],{},[119,2390,2391,2403,2416,2429,2442,2458,2471,2485,2497],{},[107,2392,2393,2395,2397,2399,2401],{},[124,2394,686],{},[124,2396,2258],{},[124,2398,689],{},[124,2400,697],{},[124,2402,689],{},[107,2404,2405,2407,2410,2412,2414],{},[124,2406,530],{},[124,2408,2409],{},"✅ 80+",[124,2411,276],{},[124,2413,276],{},[124,2415,709],{},[107,2417,2418,2420,2423,2425,2427],{},[124,2419,2225],{},[124,2421,2422],{},"✅ 一等",[124,2424,276],{},[124,2426,724],{},[124,2428,724],{},[107,2430,2431,2433,2435,2437,2440],{},[124,2432,716],{},[124,2434,276],{},[124,2436,276],{},[124,2438,2439],{},"✅ + oikb",[124,2441,724],{},[107,2443,2444,2447,2450,2453,2456],{},[124,2445,2446],{},"插件 \u002F 助手市场",[124,2448,2449],{},"✅ 丰富",[124,2451,2452],{},"300+ 助手",[124,2454,2455],{},"Tools",[124,2457,724],{},[107,2459,2460,2462,2464,2466,2469],{},[124,2461,731],{},[124,2463,276],{},[124,2465,276],{},[124,2467,2468],{},"✅ mcpo",[124,2470,724],{},[107,2472,2473,2476,2479,2481,2483],{},[124,2474,2475],{},"多用户",[124,2477,2478],{},"配 Cloud \u002F 自建",[124,2480,753],{},[124,2482,2422],{},[124,2484,753],{},[107,2486,2487,2489,2491,2493,2495],{},[124,2488,789],{},[124,2490,795],{},[124,2492,792],{},[124,2494,801],{},[124,2496,798],{},[107,2498,2499,2501,2503,2505,2507],{},[124,2500,313],{},[124,2502,322],{},[124,2504,777],{},[124,2506,322],{},[124,2508,782],{},[20,2510,328],{"id":328},[40,2512,2513,2519,2525,2531,2537,2543],{},[43,2514,2515,2518],{},[29,2516,2517],{},"Web 版数据不本地","：隐私敏感选桌面或 Docker 自托管",[43,2520,2521,2524],{},[29,2522,2523],{},"国内连海外模型走中转","：直连 OpenAI \u002F Claude 不稳，配 OpenRouter \u002F Ofox \u002F 国内中转",[43,2526,2527,2530],{},[29,2528,2529],{},"Docker 自托管暴露公网","：上反代 + HTTPS + Auth + 备份数据库",[43,2532,2533,2536],{},[29,2534,2535],{},"嵌入模型中文优化","：默认嵌入对中文一般，配 bge-m3 \u002F 硅基流动 Pro 版",[43,2538,2539,2542],{},[29,2540,2541],{},"插件市场审一遍","：第三方插件可执行代码，团队部署谨慎启用",[43,2544,2545,2548],{},[29,2546,2547],{},"同步选 Cloud vs WebDAV","：团队多端走 LobeHub Cloud；个人多设备 WebDAV 即可",[20,2550,364],{"id":363},[40,2552,2553,2556,2559,2562,2565,2568,2571,2574],{},[43,2554,2555],{},"✅ Web + 桌面双形态需求",[43,2557,2558],{},"✅ Docker 自托管 \u002F 团队共享",[43,2560,2561],{},"✅ 多模型对比 \u002F 选型",[43,2563,2564],{},"✅ 中文重度用户",[43,2566,2567],{},"✅ 助手市场 \u002F 插件生态用户",[43,2569,2570],{},"❌ 强企业 RBAC + 多租户（Open WebUI 更完善）",[43,2572,2573],{},"❌ 只要桌面 + 数据完全本地（Cherry Studio 同样优秀）",[43,2575,2576],{},"❌ 完全不会碰 Docker",[20,2578,412],{"id":412},[40,2580,2581,2585,2589,2593],{},[43,2582,2583],{},[416,2584,2101],{"href":944},[43,2586,2587],{},[416,2588,2096],{"href":2095},[43,2590,2591],{},[416,2592,882],{"href":881},[43,2594,2595],{},[416,2596,888],{"href":887},[20,2598,431],{"id":431},[191,2600,2601,2608,2615],{},[43,2602,2603,2604],{},"LobeChat GitHub 仓库（72k+ stars，MIT）",[416,2605,2606],{"href":2606,"rel":2607},"https:\u002F\u002Fgithub.com\u002Flobehub\u002Flobe-chat",[445],[43,2609,2610,2611],{},"腾讯云开发者社区 — Lobe Chat 本地化 AI 聊天终极桌面客户端（2026-01）",[416,2612,2613],{"href":2613,"rel":2614},"https:\u002F\u002Fcloud.tencent.com\u002Fdeveloper\u002Farticle\u002F2622150",[445],[43,2616,2617,2618],{},"Ofox.ai — LobeChat 完全配置指南 2026（2026-04-17）",[416,2619,2620],{"href":2620,"rel":2621},"https:\u002F\u002Fofox.ai\u002Fzh\u002Fblog\u002Flobechat-api-configuration-guide-2026",[445],{"title":455,"searchDepth":456,"depth":456,"links":2623},[2624,2625,2626,2627,2628,2629,2630,2631,2632,2633],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":2296,"depth":459,"text":2297},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Flobe-chat.webp","LobeChat 真实评测：LobeHub 团队开源 AI 聊天框架，GitHub 72k+ stars、MIT 协议。Web + 桌面（Win\u002FMac\u002FLinux\u002FDocker）双形态，支持 OpenAI \u002F Claude \u002F Gemini \u002F DeepSeek \u002F Qwen \u002F Ollama 等 80+ 模型，内置 RAG 知识库 + 插件市场 + 助手市场 + 多模型对比。",[2637,2640,2643,2646],{"q":2638,"a":2639},"Web 版 vs 桌面版 vs Docker 自托管，怎么选？","Web 版（chat.lobehub.com）最快上手但数据存 LobeHub 服务器；桌面版数据本地存、隐私好；Docker 自托管对团队 \u002F 公司部署最优，完全掌控数据。",{"q":2641,"a":2642},"支持哪些模型？","80+ 模型：OpenAI 全系列、Anthropic Claude、Google Gemini、DeepSeek、Qwen、Kimi、Moonshot、字节豆包、Groq、Together、OpenRouter、Ollama \u002F LM Studio 本地模型，以及任何 OpenAI 兼容 API。",{"q":2644,"a":2645},"多模型对比怎么用？","同一对话窗口里把消息广播给多个模型并排回答，选型 \u002F 评估特别有用——直接看 Claude 和 GPT 在同一 prompt 下的回答差异。",{"q":2647,"a":2648},"助手市场是什么？","LobeHub 维护的预设 AI 角色市场（代码审查 \u002F 翻译 \u002F 写作 \u002F 角色扮演等几百个），一键拉到本地用，省去自己写 System Prompt。",[942,476],{},[2652,483,484,485,2653],"web","docker",[2655,2658],{"plan":2284,"price":949,"features":2656,"notes":2657},"全功能 \u002F 80+ 模型 \u002F 知识库 \u002F 插件 \u002F 助手市场","MIT 协议",{"plan":2264,"price":2659,"features":2660,"notes":2661},"订阅制","云端托管 \u002F 免部署 \u002F 团队协作 \u002F 同步","chat.lobehub.com 注册即用","完全免费（MIT 开源） \u002F LobeHub Cloud 订阅",[959,2173],{"power":490,"ux":490,"price":490,"cn_support":490,"stability":489},{"title":674,"description":2635},"LobeChat - 开源 AI 聊天框架评测与部署 | AIHO",[2668,2670,2672],{"name":2669,"url":2606,"accessed":967},"LobeChat GitHub",{"name":2671,"url":2613,"accessed":967},"腾讯云开发者社区 — Lobe Chat 终极桌面客户端",{"name":2673,"url":2620,"accessed":967},"Ofox.ai — LobeChat 完全配置指南 2026","tools\u002Fcoding\u002Flocal\u002Flobe-chat","现代设计的开源 AI 聊天框架——Web + 桌面双形态、72k+ stars、多模型 + 知识库 + 插件市场",[470,2652,500,975,977,2677,2678,978],"plugin","mcp","颜值与功能双优的多模型 AI 聊天客户端。要 Web + 桌面双形态、自托管 Docker、多模型对比、丰富助手市场——LobeChat 是综合最强；纯桌面体验 Cherry Studio 同样优秀。","https:\u002F\u002Flobehub.com","zXTKiUcNATcrx3M_OoChdSfMOK8DHdQfiQ4Y7Ebo0_0",{"id":2683,"title":230,"alternatives":2684,"api_compatible":15,"body":2685,"category":470,"chinese_friendly":456,"cover":3102,"description":3103,"domestic":473,"extension":474,"faq":3104,"free":473,"github":15,"languages":3117,"lastVerified":15,"meta":3118,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":881,"pillar":481,"platforms":3119,"priceTable":3120,"pricing":3124,"published":957,"relatedPlaybooks":3125,"relatedReviews":15,"score":3126,"self_host":479,"seo":3127,"seoTitle":3128,"slug":14,"sources":3129,"stem":3135,"suitable":15,"tagline":3136,"tags":3137,"updated":967,"verdict":3142,"website":3143,"__hash__":3144},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Follama.md",[12,511,13,510],{"type":17,"value":2686,"toc":3090},[2687,2689,2696,2699,2701,2769,2771,2774,2778,2782,2802,2806,2837,2839,2873,2875,2988,2990,3022,3024,3047,3049,3067,3069],[20,2688,23],{"id":22},[25,2690,2691,2692,2695],{},"Ollama 是本地 LLM 的 Daemon 事实标准——后台跑、暴露 REST API（11434）+ CLI、Modelfile 配置、GGUF 一站式。MIT 开源，跨 Win \u002F Mac \u002F Linux。0.19+ 起 Mac M 系列底层切 MLX 推理。模型库覆盖 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral 等主流开源模型，",[1152,2693,2694],{},"ollama pull"," 一键拉。",[25,2697,2698],{},"适合：给 Cursor \u002F Cline \u002F Continue \u002F Open WebUI 接本地 OpenAI 兼容 endpoint、个人 \u002F 评估 \u002F 原型、嵌入应用、自动化脚本。不适合：GUI 偏好用户（用 LM Studio）、多用户并发生产服务（用 vLLM）、模型浏览 \u002F 调参界面（用 LM Studio）。",[20,2700,38],{"id":38},[40,2702,2703,2709,2717,2723,2730,2745,2751,2757,2763],{},[43,2704,2705,2708],{},[29,2706,2707],{},"后台 Daemon","：开机自启，应用调用零延迟",[43,2710,2711,1762,2714],{},[29,2712,2713],{},"CLI",[1152,2715,2716],{},"ollama pull \u002F run \u002F list \u002F show \u002F create \u002F serve",[43,2718,2719,2722],{},[29,2720,2721],{},"Modelfile","：类 Dockerfile 注册任意 GGUF，配 SYSTEM \u002F PARAMETER \u002F TEMPLATE",[43,2724,2725,1762,2727],{},[29,2726,1031],{},[1152,2728,2729],{},"http:\u002F\u002Flocalhost:11434\u002Fv1\u002Fchat\u002Fcompletions",[43,2731,2732,1762,2735,2738,2739,2738,2742],{},[29,2733,2734],{},"原生 API",[1152,2736,2737],{},"\u002Fapi\u002Fchat","、",[1152,2740,2741],{},"\u002Fapi\u002Fgenerate",[1152,2743,2744],{},"\u002Fapi\u002Fembeddings",[43,2746,2747,2750],{},[29,2748,2749],{},"模型库","：官方注册表内置 Llama \u002F Qwen \u002F DeepSeek \u002F Gemma \u002F Mistral \u002F GPT-OSS 等",[43,2752,2753,2756],{},[29,2754,2755],{},"MLX 加速（Mac）","：0.19+ 起 M 系列自动用 MLX",[43,2758,2759,2762],{},[29,2760,2761],{},"量化","：默认 Q4_K_M、支持 Q5 \u002F Q8 \u002F FP16",[43,2764,2765,2768],{},[29,2766,2767],{},"跨平台","：Win \u002F Mac \u002F Linux 安装包，Docker 官方镜像",[20,2770,93],{"id":93},[25,2772,2773],{},"完全免费、MIT 开源、商用免费。",[20,2775,2777],{"id":2776},"实测m2-pro-qwen3-coder-7b-q4","实测（M2 Pro + Qwen3-Coder-7B Q4）",[25,2779,2780],{},[29,2781,147],{},[40,2783,2784,2790,2793,2796,2799],{},[43,2785,2786,2789],{},[1152,2787,2788],{},"ollama run qwen3-coder:7b"," 一行起飞，3 秒进交互",[43,2791,2792],{},"REST API 配 Cursor \u002F Cline \u002F Continue 几乎全工具开箱即用",[43,2794,2795],{},"Modelfile 写自定义编码助手（low temperature + system prompt + 16K context）几分钟搞定",[43,2797,2798],{},"多模型并存，按需切换，内存占用合理",[43,2800,2801],{},"Mac M 系列 MLX 后比旧 GGUF 模式快显著",[25,2803,2804],{},[29,2805,169],{},[40,2807,2808,2818,2824,2831,2834],{},[43,2809,2810,2811,2814,2815],{},"默认 ",[1152,2812,2813],{},"num_ctx"," 偏小（2048），跑长上下文要在 Modelfile 加 ",[1152,2816,2817],{},"PARAMETER num_ctx 16384",[43,2819,2820,2821],{},"模型默认走 0.0.0.0:11434 ↔ Docker 容器互访要 ",[1152,2822,2823],{},"--add-host=host.docker.internal:host-gateway",[43,2825,2826,2827,2830],{},"国内 ",[1152,2828,2829],{},"ollama.com\u002Flibrary"," 下载偶有慢，可手动 HF 下 GGUF + Modelfile 自建",[43,2832,2833],{},"多用户并发吞吐显著低于 vLLM",[43,2835,2836],{},"没有 GUI，模型浏览 \u002F 参数面板要走 LM Studio \u002F Open WebUI 配合",[20,2838,189],{"id":189},[191,2840,2841,2847,2853,2858,2864,2870],{},[43,2842,2843,2846],{},[1152,2844,2845],{},"curl -fsSL https:\u002F\u002Follama.ai\u002Finstall.sh | sh","（Mac \u002F Linux）；Windows winget",[43,2848,2849,2852],{},[1152,2850,2851],{},"ollama pull qwen3-coder:7b","（按需换模型）",[43,2854,2855,2857],{},[1152,2856,2788],{}," 直接聊",[43,2859,2860,2861],{},"应用接入：baseURL = ",[1152,2862,2863],{},"http:\u002F\u002Flocalhost:11434\u002Fv1",[43,2865,2866,2867],{},"自定义：写 Modelfile → ",[1152,2868,2869],{},"ollama create my-coder -f Modelfile",[43,2871,2872],{},"进阶：装 Open WebUI 做前端 \u002F 多人共享",[20,2874,213],{"id":213},[101,2876,2877,2891],{},[104,2878,2879],{},[107,2880,2881,2883,2885,2887,2889],{},[110,2882,222],{},[110,2884,230],{},[110,2886,227],{},[110,2888,424],{},[110,2890,1907],{},[119,2892,2893,2909,2921,2934,2947,2963,2975],{},[107,2894,2895,2897,2900,2903,2906],{},[124,2896,686],{},[124,2898,2899],{},"CLI + Daemon",[124,2901,2902],{},"GUI + Headless",[124,2904,2905],{},"Python Server",[124,2907,2908],{},"C++ 二进制",[107,2910,2911,2913,2915,2917,2919],{},[124,2912,189],{},[124,2914,2016],{},[124,2916,2016],{},[124,2918,246],{},[124,2920,243],{},[107,2922,2923,2925,2927,2930,2932],{},[124,2924,1930],{},[124,2926,2713],{},[124,2928,2929],{},"✅ GUI",[124,2931,753],{},[124,2933,753],{},[107,2935,2936,2939,2941,2943,2945],{},[124,2937,2938],{},"OpenAI 兼容",[124,2940,1965],{},[124,2942,1962],{},[124,2944,276],{},[124,2946,276],{},[107,2948,2949,2952,2955,2958,2961],{},[124,2950,2951],{},"多用户吞吐",[124,2953,2954],{},"弱（~40 tok\u002Fs）",[124,2956,2957],{},"中（50–90）",[124,2959,2960],{},"强（800–12500）",[124,2962,246],{},[107,2964,2965,2967,2969,2971,2973],{},[124,2966,1974],{},[124,2968,1979],{},[124,2970,276],{},[124,2972,1953],{},[124,2974,798],{},[107,2976,2977,2979,2981,2983,2986],{},[124,2978,1241],{},[124,2980,322],{},[124,2982,319],{},[124,2984,2985],{},"Apache 2.0",[124,2987,322],{},[20,2989,328],{"id":328},[40,2991,2992,2998,3004,3010,3016],{},[43,2993,2994,2997],{},[29,2995,2996],{},"num_ctx 一定要设","：默认 2K 太小，跑代码 \u002F 长文档要 16K+",[43,2999,3000,3003],{},[29,3001,3002],{},"Modelfile 模板别漏 TEMPLATE","：错的 chat template 会让模型输出乱码 \u002F 不停",[43,3005,3006,3009],{},[29,3007,3008],{},"KV cache 爆表 = 速度悬崖","：32B 模型 32K 上下文，KV cache 可能 12+ GB，超显存自动 offload 慢 10×",[43,3011,3012,3015],{},[29,3013,3014],{},"不要 0.0.0.0 直接对公网","：默认无鉴权，对外暴露走反代 + Bearer \u002F mTLS",[43,3017,3018,3021],{},[29,3019,3020],{},"Mac 让它自动用 MLX","：升 0.19+；不要手动强制 GGUF + Metal",[20,3023,364],{"id":363},[40,3025,3026,3029,3032,3035,3038,3041,3044],{},[43,3027,3028],{},"✅ 应用 \u002F IDE 接本地模型（Cursor \u002F Cline \u002F Continue）",[43,3030,3031],{},"✅ 个人 \u002F 评估 \u002F 脚本自动化",[43,3033,3034],{},"✅ Modelfile 自定义系统 prompt + 参数",[43,3036,3037],{},"✅ Mac M 系列 MLX 用户",[43,3039,3040],{},"❌ 多用户并发生产服务（用 vLLM）",[43,3042,3043],{},"❌ GUI 调参 \u002F 模型浏览（配 LM Studio \u002F Open WebUI）",[43,3045,3046],{},"❌ 极致单卡吞吐研究（直接 llama.cpp \u002F vLLM）",[20,3048,412],{"id":412},[40,3050,3051,3055,3059,3063],{},[43,3052,3053],{},[416,3054,876],{"href":875},[43,3056,3057],{},[416,3058,2096],{"href":2095},[43,3060,3061],{},[416,3062,2101],{"href":944},[43,3064,3065],{},[416,3066,888],{"href":887},[20,3068,431],{"id":431},[191,3070,3071,3078,3085],{},[43,3072,3073,3074],{},"Markaicode — Import GGUF Models into Ollama 2026（2026-05-15）",[416,3075,3076],{"href":3076,"rel":3077},"https:\u002F\u002Fmarkaicode.com\u002Fimport-gguf-models-ollama-guide",[445],[43,3079,3080,3081],{},"ComputingForGeeks — Ollama Models Cheat Sheet 2026 ",[416,3082,3083],{"href":3083,"rel":3084},"https:\u002F\u002Fcomputingforgeeks.com\u002Follama-models-cheat-sheet",[445],[43,3086,2128,3087],{},[416,3088,2131],{"href":2131,"rel":3089},[445],{"title":455,"searchDepth":456,"depth":456,"links":3091},[3092,3093,3094,3095,3096,3097,3098,3099,3100,3101],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":2776,"depth":459,"text":2777},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Follama.webp","Ollama 真实评测：本地 LLM 的事实标准 Daemon，CLI + REST API，模型库 + Modelfile + GGUF 一站式。0.19+ 在 Mac M 系列用 MLX 加速；OpenAI 兼容端点 11434；MIT 开源 + 跨平台。",[3105,3108,3111,3114],{"q":3106,"a":3107},"和 LM Studio 怎么选？","Ollama = Daemon + CLI，开机自启在 11434 端口跑，应用 \u002F IDE 调它最方便。LM Studio = GUI，模型浏览 \u002F 调参 \u002F 聊天体验更好。两者底层都基于 llama.cpp，Mac M 系列上都已切 MLX。",{"q":3109,"a":3110},"Modelfile 是什么？","类 Dockerfile 的模型配置：`FROM .\u002Fxxx.gguf` + PARAMETER \u002F TEMPLATE \u002F SYSTEM。把任意 GGUF 注册成本地模型供调用。`ollama create my-model -f Modelfile`。",{"q":3112,"a":3113},"OpenAI 兼容端点？","`http:\u002F\u002Flocalhost:11434\u002Fv1`。任何 OpenAI SDK 改 baseURL 即用。也可走原生 `\u002Fapi\u002Fchat`、`\u002Fapi\u002Fgenerate`。",{"q":3115,"a":3116},"并发能力？","单用户原型场景顺滑（~40 tok\u002Fs peak），多用户并发明显不如 vLLM（vLLM 的 PagedAttention + 连续批处理高 16–20×）。生产并发选 vLLM。",[476],{},[483,484,485,2653],[3121],{"plan":126,"price":949,"features":3122,"notes":3123},"完整 CLI + REST API + Modelfile + 模型库 + MIT 协议","全平台、商用免费","完全免费 + 开源（MIT）",[959,2173],{"power":489,"ux":489,"price":490,"cn_support":456,"stability":490},{"title":230,"description":3103},"Ollama 评测 2026：本地运行大模型，开源 AI 模型管理工具，私有化部署指南",[3130,3132,3134],{"name":3131,"url":3076,"accessed":967},"Markaicode — Import GGUF 2026",{"name":3133,"url":3083,"accessed":967},"ComputingForGeeks — Ollama Cheat Sheet 2026",{"name":2183,"url":2131,"accessed":967},"tools\u002Fcoding\u002Flocal\u002Follama","本地 LLM 的 Daemon——CLI + REST API 后台跑，给 Cursor \u002F Cline \u002F Open WebUI 接本地模型最低门槛",[470,3138,3139,3140,3141,502,2188,2191,978],"daemon","cli","rest-api","modelfile","本地 LLM 的 Daemon 事实标准，CLI \u002F Modelfile \u002F REST API 三件套配合最广泛。GUI 偏好用户走 LM Studio；多用户并发生产用 vLLM；其他场景几乎默认 Ollama。","https:\u002F\u002Follama.com","7WgXNX9uMzSH_c-dUkicVopZk_g1z8HaB_8FmD9fBps",{"id":3146,"title":679,"alternatives":3147,"api_compatible":15,"body":3148,"category":470,"chinese_friendly":489,"cover":3563,"description":3564,"domestic":473,"extension":474,"faq":3565,"free":473,"github":15,"languages":3578,"lastVerified":15,"meta":3579,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":2095,"pillar":481,"platforms":3580,"priceTable":3582,"pricing":3589,"published":957,"relatedPlaybooks":3590,"relatedReviews":15,"score":3591,"self_host":479,"seo":3592,"seoTitle":3593,"slug":511,"sources":3594,"stem":3601,"suitable":15,"tagline":3602,"tags":3603,"updated":967,"verdict":3607,"website":3608,"__hash__":3609},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fopen-webui.md",[510,13,14,12],{"type":17,"value":3149,"toc":3551},[3150,3152,3155,3158,3160,3222,3224,3227,3231,3235,3263,3267,3291,3293,3327,3329,3436,3438,3481,3483,3506,3508,3526,3528],[20,3151,23],{"id":22},[25,3153,3154],{},"Open WebUI（原 Ollama WebUI）是 MIT 开源、自托管 AI 平台，最常见用法是 Docker 跑起来给 Ollama 套一个 ChatGPT 风格前端。GitHub 126k+ stars、282M+ Docker pulls，事实上的本地 AI 前端首选。支持任意 OpenAI 兼容后端 + RAG 知识库 + 多用户账号 + 工具调用 + MCP-OpenAPI 代理 + 联网搜索 + 语音 + 图像生成。",[25,3156,3157],{},"适合：团队 \u002F 家庭 \u002F 公司部署一份共享、要 Web 端访问、多用户分账号、SearXNG 联网搜索、Confluence \u002F S3 \u002F GitHub 数据源同步。不适合：单人桌面体验（用 Cherry Studio）、零运维 \u002F 不愿碰 Docker。",[20,3159,38],{"id":38},[40,3161,3162,3168,3174,3180,3186,3192,3198,3204,3210,3216],{},[43,3163,3164,3167],{},[29,3165,3166],{},"多模型后端","：Ollama \u002F OpenAI \u002F vLLM \u002F Anthropic \u002F Groq \u002F LocalAI \u002F 任意 OpenAI 兼容",[43,3169,3170,3173],{},[29,3171,3172],{},"多用户 + RBAC","：注册 \u002F 邀请 \u002F 角色权限 \u002F 工作区隔离",[43,3175,3176,3179],{},[29,3177,3178],{},"RAG 知识库","：上传文档 \u002F 网址 \u002F SearXNG 联网搜索 → 向量化 → 对话引用",[43,3181,3182,3185],{},[29,3183,3184],{},"Tools \u002F Functions","：Python 写函数即扩展（联网 \u002F 计算器 \u002F 自定义 API）",[43,3187,3188,3191],{},[29,3189,3190],{},"mcpo","：MCP-to-OpenAPI 代理，任意 MCP 服务器接进来",[43,3193,3194,3197],{},[29,3195,3196],{},"oikb","：知识库同步本地文件夹 \u002F GitHub \u002F S3 \u002F Confluence 等 40+ 源",[43,3199,3200,3203],{},[29,3201,3202],{},"open-terminal \u002F cptr","：给 AI 真实终端 + 文件 + 沙箱执行",[43,3205,3206,3209],{},[29,3207,3208],{},"图像生成","：Stable Diffusion \u002F DALL·E \u002F 自托管接入",[43,3211,3212,3215],{},[29,3213,3214],{},"语音输入 \u002F TTS","：内置",[43,3217,3218,3221],{},[29,3219,3220],{},"企业 LTS","：custom branding + SLA + 长期支持版本（联系销售）",[20,3223,93],{"id":93},[25,3225,3226],{},"完全免费、MIT 开源、商用免费。Enterprise 提供品牌定制 + SLA + LTS。",[20,3228,3230],{"id":3229},"实测ubuntu-2404-ollama-后端-5-人小团队","实测（Ubuntu 24.04 + Ollama 后端 + 5 人小团队）",[25,3232,3233],{},[29,3234,147],{},[40,3236,3237,3244,3247,3254,3257,3260],{},[43,3238,3239,3240,3243],{},"单条 ",[1152,3241,3242],{},"docker run"," 五分钟上线",[43,3245,3246],{},"自带的多用户 + 角色权限省去重新搭 Auth",[43,3248,3249,3250,3253],{},"RAG 直传 30 个 PDF 后向量化顺利，对话中 ",[1152,3251,3252],{},"#知识库"," 引用准确",[43,3255,3256],{},"mcpo 把 GitHub MCP 服务器接进来，团队对话里直接 issue \u002F PR 操作",[43,3258,3259],{},"模型切换流畅，OpenAI + Ollama 并存",[43,3261,3262],{},"SearXNG 联网搜索给模型实时信息，过时知识截止问题缓解",[25,3264,3265],{},[29,3266,169],{},[40,3268,3269,3272,3278,3285,3288],{},[43,3270,3271],{},"Docker 镜像 ~1.5GB，首次拉取偏慢",[43,3273,2810,3274,3277],{},[1152,3275,3276],{},"0.0.0.0"," 公网暴露要加 HTTPS + 反代",[43,3279,3280,3281,3284],{},"嵌入模型 ",[1152,3282,3283],{},"sentence-transformers"," 中文效果一般，建议换 bge-m3",[43,3286,3287],{},"多用户共享 Ollama 时并发吞吐瓶颈在 Ollama，不在 Open WebUI（生产用 vLLM 后端）",[43,3289,3290],{},"版本升级要看 changelog，部分 minor 含 breaking 改动",[20,3292,189],{"id":189},[191,3294,3295,3301,3308,3311,3314,3317,3320],{},[43,3296,3297,3298],{},"装 Docker → ",[1152,3299,3300],{},"docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:\u002Fapp\u002Fbackend\u002Fdata --name open-webui --restart always ghcr.io\u002Fopen-webui\u002Fopen-webui:main",[43,3302,3303,3304,3307],{},"浏览器开 ",[1152,3305,3306],{},"http:\u002F\u002Flocalhost:3000"," → 注册第一个账号（管理员）",[43,3309,3310],{},"设置 → Connections → 连接 Ollama \u002F 加 OpenAI Key",[43,3312,3313],{},"Models → Pull \u002F Discover 模型",[43,3315,3316],{},"Workspaces → 建知识库 → 上传文档",[43,3318,3319],{},"Tools → 启用 \u002F 写自定义函数",[43,3321,3322,3323,3326],{},"生产部署：Nginx 反代 + Let's Encrypt + 备份 ",[1152,3324,3325],{},"\u002Fapp\u002Fbackend\u002Fdata"," volume",[20,3328,213],{"id":213},[101,3330,3331,3345],{},[104,3332,3333],{},[107,3334,3335,3337,3339,3341,3343],{},[110,3336,222],{},[110,3338,679],{},[110,3340,674],{},[110,3342,233],{},[110,3344,227],{},[119,3346,3347,3359,3371,3385,3398,3412,3424],{},[107,3348,3349,3351,3353,3355,3357],{},[124,3350,686],{},[124,3352,697],{},[124,3354,692],{},[124,3356,689],{},[124,3358,689],{},[107,3360,3361,3363,3365,3367,3369],{},[124,3362,2475],{},[124,3364,2422],{},[124,3366,276],{},[124,3368,753],{},[124,3370,753],{},[107,3372,3373,3376,3379,3381,3383],{},[124,3374,3375],{},"RAG",[124,3377,3378],{},"✅ 强 + oikb",[124,3380,276],{},[124,3382,276],{},[124,3384,724],{},[107,3386,3387,3390,3392,3394,3396],{},[124,3388,3389],{},"工具 \u002F MCP",[124,3391,2468],{},[124,3393,276],{},[124,3395,276],{},[124,3397,724],{},[107,3399,3400,3403,3406,3408,3410],{},[124,3401,3402],{},"自托管",[124,3404,3405],{},"✅ Docker \u002F K8s",[124,3407,750],{},[124,3409,753],{},[124,3411,753],{},[107,3413,3414,3416,3418,3420,3422],{},[124,3415,789],{},[124,3417,801],{},[124,3419,795],{},[124,3421,792],{},[124,3423,798],{},[107,3425,3426,3428,3430,3432,3434],{},[124,3427,313],{},[124,3429,322],{},[124,3431,322],{},[124,3433,777],{},[124,3435,782],{},[20,3437,328],{"id":328},[40,3439,3440,3446,3454,3463,3469,3475],{},[43,3441,3442,3445],{},[29,3443,3444],{},"不要裸 0.0.0.0 + HTTP 暴露公网","：默认无 HTTPS，必上反代 + 强密码 + 速率限制",[43,3447,3448,3453],{},[29,3449,3450,3451,3326],{},"备份 ",[1152,3452,3325],{},"：知识库 \u002F 用户 \u002F 对话全在里面",[43,3455,3456,3459,3460,3462],{},[29,3457,3458],{},"中文 RAG 换嵌入模型","：默认 ",[1152,3461,3283],{}," 中文一般，配 bge-m3 或硅基流动嵌入 API",[43,3464,3465,3468],{},[29,3466,3467],{},"mcpo 工具范围谨慎","：MCP 给 AI 真实能力，第三方服务器审一遍",[43,3470,3471,3474],{},[29,3472,3473],{},"后端吞吐看 Ollama","：5+ 并发上 vLLM 后端，Ollama 单 worker 会排队",[43,3476,3477,3480],{},[29,3478,3479],{},"升级前看 changelog","：weekly 更新，偶有 breaking",[20,3482,364],{"id":363},[40,3484,3485,3488,3491,3494,3497,3500,3503],{},[43,3486,3487],{},"✅ 团队 \u002F 家庭 \u002F 公司多人共享 AI 平台",[43,3489,3490],{},"✅ 要 Web 端访问 \u002F 移动端兼容",[43,3492,3493],{},"✅ 自托管 \u002F 完全控制数据",[43,3495,3496],{},"✅ MCP \u002F 工具调用刚需",[43,3498,3499],{},"❌ 单人桌面体验（用 Cherry Studio）",[43,3501,3502],{},"❌ 零运维 \u002F 不愿碰 Docker",[43,3504,3505],{},"❌ iOS 原生 App 主力",[20,3507,412],{"id":412},[40,3509,3510,3514,3518,3522],{},[43,3511,3512],{},[416,3513,870],{"href":869},[43,3515,3516],{},[416,3517,2101],{"href":944},[43,3519,3520],{},[416,3521,882],{"href":881},[43,3523,3524],{},[416,3525,888],{"href":887},[20,3527,431],{"id":431},[191,3529,3530,3537,3544],{},[43,3531,3532,3533],{},"Open WebUI 官方文档 ",[416,3534,3535],{"href":3535,"rel":3536},"https:\u002F\u002Fdocs.openwebui.com\u002F",[445],[43,3538,3539,3540],{},"Local AI Master — Open WebUI Setup Guide 2026 ",[416,3541,3542],{"href":3542,"rel":3543},"https:\u002F\u002Flocalaimaster.com\u002Fblog\u002Fopen-webui-setup-guide",[445],[43,3545,3546,3547],{},"AIToolDiscovery — Set Up Open-WebUI with Ollama 2026 ",[416,3548,3549],{"href":3549,"rel":3550},"https:\u002F\u002Fwww.aitooldiscovery.com\u002Fhow-to\u002Fsetup-open-webui-ollama",[445],{"title":455,"searchDepth":456,"depth":456,"links":3552},[3553,3554,3555,3556,3557,3558,3559,3560,3561,3562],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":3229,"depth":459,"text":3230},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Fopen-webui.webp","Open WebUI 2026 真实评测：MIT 开源、自托管 ChatGPT 替代和 Ollama Web 前端。支持 Docker 一行部署、Ollama\u002FOpenAI\u002FvLLM 多后端、RAG 知识库、多用户、联网搜索、工具调用和 MCP-to-OpenAPI，适合团队私有 AI 平台。",[3566,3569,3572,3575],{"q":3567,"a":3568},"Docker 一行命令真的够用吗？","够。`docker run -d -p 3000:8080 --add-host=host.docker.internal:host-gateway -v open-webui:\u002Fapp\u002Fbackend\u002Fdata --name open-webui --restart always ghcr.io\u002Fopen-webui\u002Fopen-webui:main`，5 分钟可上线、能多人注册、能接 Ollama \u002F OpenAI。生产再加反代 + HTTPS + 备份。",{"q":3570,"a":3571},"支持哪些模型后端？","Ollama（首选）+ 任何 OpenAI 兼容 endpoint：OpenAI 官方 \u002F Anthropic（OpenAI 兼容代理）\u002F vLLM \u002F Groq \u002F LocalAI \u002F 自建 baseURL。可同时配多个，对话中切换。",{"q":3573,"a":3574},"RAG \u002F 知识库怎么做？","内置：上传 PDF \u002F DOCX \u002F TXT、网址抓取、SearXNG 联网搜索 → 自动向量化 → 在对话中 `#` 引用知识库。配套 oikb 项目可同步本地文件夹 \u002F GitHub \u002F S3 \u002F Confluence 等 40+ 数据源。",{"q":3576,"a":3577},"MCP 怎么接？","通过 mcpo（官方的 MCP-to-OpenAPI 代理）把任意 MCP 服务器暴露成 OpenAPI 工具，再在 Open WebUI 注册即可。无需写 glue code。",[476,942],{},[2653,485,484,483,3581],"kubernetes",[3583,3585],{"plan":126,"price":949,"features":3584,"notes":2657},"全功能 \u002F 多用户 \u002F RAG \u002F Tools \u002F 联网搜索 \u002F MCP-OpenAPI 代理 \u002F Docker \u002F K8s",{"plan":581,"price":3586,"features":3587,"notes":3588},"咨询","Custom branding \u002F SLA \u002F LTS 长期支持版本","邮件官方","完全免费（MIT 开源） \u002F Enterprise SLA 联系",[959,2173],{"power":490,"ux":489,"price":490,"cn_support":489,"stability":490},{"title":679,"description":3564},"Open WebUI 评测 2026：自托管 ChatGPT 替代，Ollama 前端部署指南",[3595,3597,3599],{"name":3596,"url":3535,"accessed":967},"Open WebUI 官方文档",{"name":3598,"url":3542,"accessed":967},"Local AI Master — Open WebUI Setup Guide 2026",{"name":3600,"url":3549,"accessed":967},"AIToolDiscovery — Open-WebUI with Ollama 2026","tools\u002Fcoding\u002Flocal\u002Fopen-webui","自托管的 ChatGPT 替代：Ollama \u002F OpenAI 兼容、多用户、RAG、126k+ GitHub stars",[470,3604,2653,977,3605,3606,978],"self-host","multi-user","ollama","自托管多用户 AI 前端的事实标准。团队 \u002F 家庭 \u002F 公司部署一份共享，多模型聚合 + RAG + 工具调用全有。单机 \u002F 桌面体验首选 Cherry Studio \u002F LobeChat。","https:\u002F\u002Fdocs.openwebui.com","JCKn_X0aojpl94LKJcqlbs0XSTAxv8S8JgKy70WtsO0",{"id":3611,"title":424,"alternatives":3612,"api_compatible":15,"body":3613,"category":470,"chinese_friendly":459,"cover":4089,"description":4090,"domestic":473,"extension":474,"faq":15,"free":473,"github":4074,"languages":4091,"lastVerified":477,"meta":4092,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":4093,"pillar":481,"platforms":4094,"priceTable":15,"pricing":4095,"published":487,"relatedPlaybooks":15,"relatedReviews":15,"score":4096,"self_host":473,"seo":4097,"seoTitle":4098,"slug":4099,"sources":4100,"stem":4103,"suitable":15,"tagline":4104,"tags":4105,"updated":477,"verdict":4110,"website":4068,"__hash__":4111},"tools\u002Ftools\u002Fcoding\u002Flocal\u002Fvllm.md",[14,12],{"type":17,"value":3614,"toc":4076},[3615,3617,3620,3623,3625,3685,3687,3690,3692,3700,3704,3727,3731,3763,3765,3805,3807,3936,3938,3992,3994,4020,4022,4028,4034,4040,4046,4048,4056,4058,4062],[20,3616,23],{"id":22},[25,3618,3619],{},"vLLM 是当前开源生态吞吐量最高的 LLM 推理引擎，由 UC Berkeley 团队开发，核心创新 PagedAttention 把 KV cache 当虚拟内存管，配合连续批处理（continuous batching）把 GPU 利用率从传统推理的 30-40% 拉到 70-80%+。Apache 2.0 协议，纯 Python + CUDA，部署在 Linux + NVIDIA GPU。",[25,3621,3622],{},"适合：需要对外提供 LLM API 服务、多用户并发、追求最大吞吐和最低延迟的工程团队，以及跑大规模 batch 离线推理的研究场景。不适合：单用户本地原型（用 Ollama 更轻量）、Mac M 系列（vLLM 对 Metal 支持有限）、没有 NVIDIA GPU 的环境、不想碰 Linux + CUDA 驱动的小团队。",[20,3624,38],{"id":38},[40,3626,3627,3633,3639,3645,3658,3664,3673,3679],{},[43,3628,3629,3632],{},[29,3630,3631],{},"PagedAttention","：借鉴操作系统虚拟内存的分页机制管理 KV cache，消除碎片化，显存利用率提升 2-4 倍",[43,3634,3635,3638],{},[29,3636,3637],{},"连续批处理（Continuous Batching）","：请求动态插入 \u002F 弹出，不需要等整批完成，GPU 闲置接近为零",[43,3640,3641,3644],{},[29,3642,3643],{},"高并发吞吐","：单 A100 跑 Llama-3-8B 可达 800-12500 tok\u002Fs（取决于 batch size），比 Hugging Face Transformers 高 14-24 倍",[43,3646,3647,3649,3650,3653,3654,3657],{},[29,3648,1031],{},"：内置 ",[1152,3651,3652],{},"--api-server","，端点 ",[1152,3655,3656],{},"\u002Fv1\u002Fchat\u002Fcompletions"," 直接替换 OpenAI SDK 的 baseURL 即用",[43,3659,3660,3663],{},[29,3661,3662],{},"量化支持","：AWQ、GPTQ、FP8（H100\u002FAda）、INT8 KV cache，显存减半吞吐不掉",[43,3665,3666,1762,3669,3672],{},[29,3667,3668],{},"张量并行（Tensor Parallelism）",[1152,3670,3671],{},"--tensor-parallel-size N"," 多卡切分，支持多 GPU 推理大模型",[43,3674,3675,3678],{},[29,3676,3677],{},"分布式部署","：Ray 集群多节点推理，支持 pipeline parallelism",[43,3680,3681,3684],{},[29,3682,3683],{},"LoRA 多租户","：同时加载多个 LoRA adapter，单服务多模型，按请求路由",[20,3686,93],{"id":93},[25,3688,3689],{},"完全免费，Apache 2.0 开源，商用无限制。成本在于 GPU 硬件：一张 A100 80GB 云端约 $2-4\u002F小时（按需），跑 70B 模型需 2-4 张。自建机房摊薄后更便宜。",[20,3691,139],{"id":138},[95,3693,3694],{},[25,3695,144,3696,3699],{},[29,3697,3698],{},"环境（撰写时参考）","：4× A100 80GB + Llama-3-70B-Instruct（FP16），vLLM 0.6.x 系列（最新稳定版请以 vllm.ai 为准）。",[25,3701,3702],{},[29,3703,147],{},[40,3705,3706,3712,3715,3718,3721,3724],{},[43,3707,3708,3711],{},[1152,3709,3710],{},"vllm serve meta-llama\u002FMeta-Llama-3-70B-Instruct --tensor-parallel-size 4"," 一行拉起，4 卡自动切分",[43,3713,3714],{},"并发 64 用户，平均延迟 1.2s，吞吐稳定在 3200 tok\u002Fs，GPU 利用率 75-85%",[43,3716,3717],{},"同样硬件跑 HF Transformers + 默认 batching，吞吐仅 ~200 tok\u002Fs，差距 16 倍",[43,3719,3720],{},"AWQ 量化版 70B 单卡 A100 即可跑，吞吐只掉 15-20%，显存从 140GB 降到 40GB",[43,3722,3723],{},"OpenAI 兼容端点接 Cursor \u002F Dify \u002F FastGPT 零改动",[43,3725,3726],{},"连续批处理下短请求和长请求混合调度公平，没有长尾饿死",[25,3728,3729],{},[29,3730,169],{},[40,3732,3733,3740,3747,3750,3757],{},[43,3734,3735,3736,3739],{},"第一次启动要编译 CUDA kernel，冷启动 3-5 分钟，加 ",[1152,3737,3738],{},"--enforce-eager"," 可跳过但掉速 20%",[43,3741,3742,3743,3746],{},"KV cache 默认占 90% 显存，跑长上下文（32K+）要手动调 ",[1152,3744,3745],{},"--gpu-memory-utilization 0.85"," 留余量",[43,3748,3749],{},"旧版本对 Qwen2.5-VL 等多模态模型支持不稳定，偶发 OOM，建议查阅官方 issue 选择适配版本",[43,3751,3752,3753,3756],{},"国内 HuggingFace 下载模型慢，配 ",[1152,3754,3755],{},"HF_ENDPOINT=https:\u002F\u002Fhf-mirror.com"," 或预下载到本地",[43,3758,3759,3762],{},[1152,3760,3761],{},"--max-model-len"," 必须设，否则默认按模型最大上下文分配，32B 模型 128K 上下文会直接 OOM",[20,3764,189],{"id":189},[191,3766,3767,3773,3779,3785,3792,3798],{},[43,3768,3769,3770],{},"环境准备：Linux + NVIDIA GPU（compute capability ≥ 7.0）+ CUDA 12.1+，",[1152,3771,3772],{},"pip install vllm",[43,3774,3775,3776],{},"拉起服务：",[1152,3777,3778],{},"vllm serve meta-llama\u002FMeta-Llama-3-8B-Instruct --port 8000",[43,3780,3781,3782],{},"测试调用：",[1152,3783,3784],{},"curl http:\u002F\u002Flocalhost:8000\u002Fv1\u002Fchat\u002Fcompletions -H \"Content-Type: application\u002Fjson\" -d '{\"model\":\"meta-llama\u002FMeta-Llama-3-8B-Instruct\",\"messages\":[{\"role\":\"user\",\"content\":\"hi\"}]}'",[43,3786,3787,3788,3791],{},"多卡并行：加 ",[1152,3789,3790],{},"--tensor-parallel-size 4","（卡数）",[43,3793,3794,3795],{},"量化部署：",[1152,3796,3797],{},"vllm serve TheBloke\u002FLlama-2-13B-AWQ --quantization awq",[43,3799,3800,3801,3804],{},"接入应用：任何 OpenAI SDK 改 ",[1152,3802,3803],{},"base_url=http:\u002F\u002Flocalhost:8000\u002Fv1"," 即用",[20,3806,213],{"id":213},[101,3808,3809,3825],{},[104,3810,3811],{},[107,3812,3813,3815,3817,3819,3822],{},[110,3814,222],{},[110,3816,424],{},[110,3818,230],{},[110,3820,3821],{},"TGI (HF)",[110,3823,3824],{},"TensorRT-LLM",[119,3826,3827,3844,3857,3870,3883,3896,3908,3923],{},[107,3828,3829,3832,3835,3838,3841],{},[124,3830,3831],{},"吞吐（A100 8B）",[124,3833,3834],{},"~800-12500 tok\u002Fs",[124,3836,3837],{},"~40 tok\u002Fs",[124,3839,3840],{},"~500 tok\u002Fs",[124,3842,3843],{},"~10000 tok\u002Fs",[107,3845,3846,3849,3851,3853,3855],{},[124,3847,3848],{},"上手门槛",[124,3850,246],{},[124,3852,2016],{},[124,3854,246],{},[124,3856,243],{},[107,3858,3859,3861,3863,3865,3868],{},[124,3860,3631],{},[124,3862,276],{},[124,3864,290],{},[124,3866,3867],{},"✅ (v0.7+)",[124,3869,290],{},[107,3871,3872,3875,3877,3879,3881],{},[124,3873,3874],{},"连续批处理",[124,3876,276],{},[124,3878,290],{},[124,3880,276],{},[124,3882,276],{},[107,3884,3885,3887,3889,3891,3893],{},[124,3886,2938],{},[124,3888,276],{},[124,3890,276],{},[124,3892,276],{},[124,3894,3895],{},"需封装",[107,3897,3898,3900,3902,3904,3906],{},[124,3899,560],{},[124,3901,1953],{},[124,3903,276],{},[124,3905,276],{},[124,3907,1953],{},[107,3909,3910,3913,3916,3919,3921],{},[124,3911,3912],{},"Mac 支持",[124,3914,3915],{},"❌ 有限",[124,3917,3918],{},"✅ MLX",[124,3920,290],{},[124,3922,290],{},[107,3924,3925,3927,3929,3931,3934],{},[124,3926,313],{},[124,3928,2985],{},[124,3930,322],{},[124,3932,3933],{},"HFOIL",[124,3935,2985],{},[20,3937,328],{"id":328},[40,3939,3940,3949,3957,3971,3977,3986],{},[43,3941,3942,3945,3946,3948],{},[29,3943,3944],{},"冷启动慢不是 bug","：首次编译 CUDA kernel 需要几分钟，生产环境用 Docker 镜像预编译或加 ",[1152,3947,3738],{},"（牺牲 15-20% 性能换即时启动）",[43,3950,3951,3956],{},[29,3952,3953,3955],{},[1152,3954,3761],{}," 必设","：不设会按模型最大上下文预分配 KV cache，小显存直接 OOM",[43,3958,3959,3962,3963,3966,3967,3970],{},[29,3960,3961],{},"量化模型要匹配版本","：AWQ 模型必须用 ",[1152,3964,3965],{},"--quantization awq","，GPTQ 用 ",[1152,3968,3969],{},"--quantization gptq","，混用会报错或精度崩",[43,3972,3973,3976],{},[29,3974,3975],{},"不要在 Mac 上用 vLLM 跑生产","：Metal 后端是实验性的，性能远不如 CPU，Mac 本地推理用 Ollama \u002F MLX",[43,3978,3979,3982,3983,3985],{},[29,3980,3981],{},"监控 GPU 显存碎片","：长跑后偶发显存碎片导致新请求 OOM，加 ",[1152,3984,3745],{}," 留 buffer 或定期重启",[43,3987,3988,3991],{},[29,3989,3990],{},"多模态模型看版本","：不同版本对 VLM 支持差异较大，新模型先查官方 issue 选适配版本",[20,3993,364],{"id":363},[40,3995,3996,3999,4002,4005,4008,4011,4014,4017],{},[43,3997,3998],{},"✅ 生产级 LLM API 服务（多用户并发、高吞吐）",[43,4000,4001],{},"✅ 大规模离线 batch 推理（数据标注、合成数据生成）",[43,4003,4004],{},"✅ 需要最低成本跑大模型（量化 + 单卡部署 70B）",[43,4006,4007],{},"✅ 有 NVIDIA GPU + Linux 运维能力的工程团队",[43,4009,4010],{},"❌ 单用户本地原型 \u002F 个人开发（用 Ollama，0 配置）",[43,4012,4013],{},"❌ Mac M 系列用户（Metal 支持有限，用 Ollama + MLX）",[43,4015,4016],{},"❌ 没有 GPU 的环境（vLLM 的 CPU 后端性能极差）",[43,4018,4019],{},"❌ 多模态 \u002F 语音模型生产部署（支持不稳定，看具体版本）",[20,4021,391],{"id":390},[25,4023,4024,4027],{},[29,4025,4026],{},"Q: vLLM 和 Ollama 怎么选？","\nA: Ollama 是 Daemon + CLI，单用户原型极简；vLLM 是推理服务器，多用户并发吞吐高 16-20 倍。个人用 Ollama，对外提供服务用 vLLM。",[25,4029,4030,4033],{},[29,4031,4032],{},"Q: 单卡能跑 70B 吗？","\nA: 可以。用 AWQ\u002FGPTQ 4-bit 量化，70B 约需 35-40GB 显存，A100 80GB 或 2×A100 40GB 张量并行。FP16 则需 140GB（2×A100 80GB）。",[25,4035,4036,4039],{},[29,4037,4038],{},"Q: 和 TensorRT-LLM 比谁快？","\nA: TensorRT-LLM 在极致优化下略快（5-15%），但需要编译 engine、调试周期长、模型适配少。vLLM 灵活性和生态好得多，综合性价比更高。",[25,4041,4042,4045],{},[29,4043,4044],{},"Q: 支持 AMD GPU 吗？","\nA: 部分支持。0.5+ 起 ROCm 后端可用，但稳定性、性能、生态都远不如 NVIDIA CUDA。生产环境仍建议 NVIDIA。",[20,4047,412],{"id":412},[25,4049,4050,420,4052,420,4054],{},[416,4051,10],{"href":1332},[416,4053,419],{"href":418},[416,4055,428],{"href":427},[20,4057,431],{"id":431},[95,4059,4060],{},[25,4061,436],{},[40,4063,4064,4070],{},[43,4065,4066],{},[416,4067,446],{"href":4068,"rel":4069},"https:\u002F\u002Fvllm.ai",[445],[43,4071,4072],{},[416,4073,453],{"href":4074,"rel":4075},"https:\u002F\u002Fgithub.com\u002Fvllm-project\u002Fvllm",[445],{"title":455,"searchDepth":456,"depth":456,"links":4077},[4078,4079,4080,4081,4082,4083,4084,4085,4086,4087,4088],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":138,"depth":459,"text":139},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":390,"depth":459,"text":391},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},"\u002Fimg\u002Ftools\u002Fvllm.webp","vLLM 真实评测：开源高吞吐 LLM 推理引擎（Apache 2.0 协议），核心创新 PagedAttention + 连续批处理，显著提升 GPU 利用率和推理吞吐。适合需要生产级高并发 LLM 服务部署、追求最大吞吐量的工程团队。",[476],{},"\u002Ftools\u002Fcoding\u002Flocal\u002Fvllm",[485],"Free \u002F 开源（Apache 2.0）",{"power":490,"ux":456,"price":490,"cn_support":459,"stability":489},{"title":424,"description":4090},"vLLM - 高吞吐 LLM 推理引擎评测与部署 | AIHO","coding\u002Flocal\u002Fvllm",[4101,4102],{"title":446,"url":4068},{"title":453,"url":4074},"tools\u002Fcoding\u002Flocal\u002Fvllm","高吞吐 LLM 推理引擎，PagedAttention 连续批处理",[470,4106,4107,4108,4109],"inference-engine","production","paged-attention","gpu","生产级高并发 LLM 服务的首选推理引擎，PagedAttention + 连续批处理把单卡吞吐拉到极致；单用户原型和 Mac 本地玩用 Ollama \u002F LM Studio 更省心。","njZ35H6xUwwjCk237XJvzfgRSW0kRrEEdoBySqYygng",[],[],[4115,4454,4804],{"id":508,"title":233,"alternatives":4116,"api_compatible":15,"body":4117,"category":470,"chinese_friendly":490,"cover":926,"description":927,"domestic":473,"extension":474,"faq":4435,"free":473,"github":15,"languages":4440,"lastVerified":15,"meta":4441,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":944,"pillar":481,"platforms":4442,"priceTable":4443,"pricing":956,"published":957,"relatedPlaybooks":4446,"relatedReviews":15,"score":4447,"self_host":479,"seo":4448,"seoTitle":963,"slug":13,"sources":4449,"stem":972,"suitable":15,"tagline":973,"tags":4453,"updated":967,"verdict":980,"website":981,"__hash__":982},[510,12,14,511],{"type":17,"value":4118,"toc":4423},[4119,4121,4123,4125,4127,4157,4159,4169,4173,4175,4179,4191,4195,4207,4209,4226,4228,4342,4344,4366,4368,4384,4386,4404,4406],[20,4120,23],{"id":22},[25,4122,518],{},[25,4124,521],{},[20,4126,38],{"id":38},[40,4128,4129,4133,4137,4141,4145,4149,4153],{},[43,4130,4131,531],{},[29,4132,530],{},[43,4134,4135,537],{},[29,4136,536],{},[43,4138,4139,543],{},[29,4140,542],{},[43,4142,4143,549],{},[29,4144,548],{},[43,4146,4147,555],{},[29,4148,554],{},[43,4150,4151,561],{},[29,4152,560],{},[43,4154,4155,567],{},[29,4156,566],{},[20,4158,93],{"id":93},[40,4160,4161,4165],{},[43,4162,4163,576],{},[29,4164,126],{},[43,4166,4167,582],{},[29,4168,581],{},[95,4170,4171],{},[25,4172,587],{},[20,4174,591],{"id":590},[25,4176,4177],{},[29,4178,147],{},[40,4180,4181,4183,4185,4187,4189],{},[43,4182,600],{},[43,4184,603],{},[43,4186,606],{},[43,4188,609],{},[43,4190,612],{},[25,4192,4193],{},[29,4194,169],{},[40,4196,4197,4199,4201,4203,4205],{},[43,4198,621],{},[43,4200,624],{},[43,4202,627],{},[43,4204,630],{},[43,4206,633],{},[20,4208,189],{"id":189},[191,4210,4211,4213,4215,4220,4222,4224],{},[43,4212,640],{},[43,4214,643],{},[43,4216,646,4217],{},[416,4218,649],{"href":649,"rel":4219},[445],[43,4221,653],{},[43,4223,656],{},[43,4225,659],{},[20,4227,213],{"id":213},[101,4229,4230,4244],{},[104,4231,4232],{},[107,4233,4234,4236,4238,4240,4242],{},[110,4235,222],{},[110,4237,233],{},[110,4239,674],{},[110,4241,227],{},[110,4243,679],{},[119,4245,4246,4258,4270,4282,4294,4306,4318,4330],{},[107,4247,4248,4250,4252,4254,4256],{},[124,4249,686],{},[124,4251,689],{},[124,4253,692],{},[124,4255,689],{},[124,4257,697],{},[107,4259,4260,4262,4264,4266,4268],{},[124,4261,530],{},[124,4263,704],{},[124,4265,704],{},[124,4267,709],{},[124,4269,704],{},[107,4271,4272,4274,4276,4278,4280],{},[124,4273,716],{},[124,4275,719],{},[124,4277,719],{},[124,4279,724],{},[124,4281,276],{},[107,4283,4284,4286,4288,4290,4292],{},[124,4285,731],{},[124,4287,276],{},[124,4289,276],{},[124,4291,724],{},[124,4293,276],{},[107,4295,4296,4298,4300,4302,4304],{},[124,4297,744],{},[124,4299,747],{},[124,4301,750],{},[124,4303,753],{},[124,4305,750],{},[107,4307,4308,4310,4312,4314,4316],{},[124,4309,760],{},[124,4311,763],{},[124,4313,763],{},[124,4315,768],{},[124,4317,768],{},[107,4319,4320,4322,4324,4326,4328],{},[124,4321,313],{},[124,4323,777],{},[124,4325,322],{},[124,4327,782],{},[124,4329,322],{},[107,4331,4332,4334,4336,4338,4340],{},[124,4333,789],{},[124,4335,792],{},[124,4337,795],{},[124,4339,798],{},[124,4341,801],{},[20,4343,328],{"id":328},[40,4345,4346,4350,4354,4358,4362],{},[43,4347,4348,811],{},[29,4349,810],{},[43,4351,4352,817],{},[29,4353,816],{},[43,4355,4356,823],{},[29,4357,822],{},[43,4359,4360,829],{},[29,4361,828],{},[43,4363,4364,835],{},[29,4365,834],{},[20,4367,364],{"id":363},[40,4369,4370,4372,4374,4376,4378,4380,4382],{},[43,4371,842],{},[43,4373,845],{},[43,4375,848],{},[43,4377,851],{},[43,4379,854],{},[43,4381,857],{},[43,4383,860],{},[20,4385,412],{"id":412},[40,4387,4388,4392,4396,4400],{},[43,4389,4390],{},[416,4391,870],{"href":869},[43,4393,4394],{},[416,4395,876],{"href":875},[43,4397,4398],{},[416,4399,882],{"href":881},[43,4401,4402],{},[416,4403,888],{"href":887},[20,4405,431],{"id":431},[191,4407,4408,4413,4418],{},[43,4409,895,4410],{},[416,4411,898],{"href":898,"rel":4412},[445],[43,4414,902,4415],{},[416,4416,905],{"href":905,"rel":4417},[445],[43,4419,909,4420],{},[416,4421,912],{"href":912,"rel":4422},[445],{"title":455,"searchDepth":456,"depth":456,"links":4424},[4425,4426,4427,4428,4429,4430,4431,4432,4433,4434],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":590,"depth":459,"text":591},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},[4436,4437,4438,4439],{"q":930,"a":931},{"q":933,"a":934},{"q":936,"a":937},{"q":939,"a":940},[942,476],{},[483,484,485,946],[4444,4445],{"plan":126,"price":949,"features":950,"notes":951},{"plan":581,"price":953,"features":954,"notes":955},[959,960],{"power":489,"ux":490,"price":490,"cn_support":490,"stability":489},{"title":233,"description":927},[4450,4451,4452],{"name":966,"url":898,"accessed":967},{"name":969,"url":905,"accessed":967},{"name":971,"url":912,"accessed":967},[470,500,975,976,977,978,979],{"id":1719,"title":227,"alternatives":4455,"api_compatible":15,"body":4456,"category":470,"chinese_friendly":456,"cover":2145,"description":2146,"domestic":473,"extension":474,"faq":4785,"free":473,"github":15,"languages":4790,"lastVerified":15,"meta":4791,"models":15,"navigation":479,"notSuitable":15,"opensource":473,"path":875,"pillar":481,"platforms":4792,"priceTable":4793,"pricing":2171,"published":957,"relatedPlaybooks":4796,"relatedReviews":15,"score":4797,"self_host":479,"seo":4798,"seoTitle":2176,"slug":12,"sources":4799,"stem":2184,"suitable":15,"tagline":2185,"tags":4803,"updated":967,"verdict":2192,"website":2193,"__hash__":2194},[14,511,13,510],{"type":17,"value":4457,"toc":4773},[4458,4460,4464,4466,4468,4506,4508,4518,4522,4524,4528,4540,4544,4556,4558,4576,4578,4692,4694,4716,4718,4734,4736,4754,4756],[20,4459,23],{"id":22},[25,4461,1727,4462,1731],{},[1152,4463,1730],{},[25,4465,1734],{},[20,4467,38],{"id":38},[40,4469,4470,4474,4478,4482,4488,4494,4498,4502],{},[43,4471,4472,1744],{},[29,4473,1743],{},[43,4475,4476,1750],{},[29,4477,1749],{},[43,4479,4480,1756],{},[29,4481,1755],{},[43,4483,4484,1762,4486,1766],{},[29,4485,1761],{},[1152,4487,1765],{},[43,4489,4490,1762,4492,1775],{},[29,4491,1771],{},[1152,4493,1774],{},[43,4495,4496,1781],{},[29,4497,1780],{},[43,4499,4500,1787],{},[29,4501,1786],{},[43,4503,4504,1793],{},[29,4505,1792],{},[20,4507,93],{"id":93},[40,4509,4510,4514],{},[43,4511,4512,1803],{},[29,4513,1802],{},[43,4515,4516,1809],{},[29,4517,1808],{},[95,4519,4520],{},[25,4521,1814],{},[20,4523,1818],{"id":1817},[25,4525,4526],{},[29,4527,147],{},[40,4529,4530,4532,4534,4536,4538],{},[43,4531,1827],{},[43,4533,1830],{},[43,4535,1833],{},[43,4537,1836],{},[43,4539,1839],{},[25,4541,4542],{},[29,4543,169],{},[40,4545,4546,4548,4550,4552,4554],{},[43,4547,1848],{},[43,4549,1851],{},[43,4551,1854],{},[43,4553,1857],{},[43,4555,1860],{},[20,4557,189],{"id":189},[191,4559,4560,4562,4564,4566,4568,4572],{},[43,4561,1867],{},[43,4563,1870],{},[43,4565,1873],{},[43,4567,1876],{},[43,4569,1879,4570,1883],{},[1152,4571,1882],{},[43,4573,1886,4574],{},[1152,4575,1774],{},[20,4577,213],{"id":213},[101,4579,4580,4594],{},[104,4581,4582],{},[107,4583,4584,4586,4588,4590,4592],{},[110,4585,222],{},[110,4587,227],{},[110,4589,230],{},[110,4591,679],{},[110,4593,1907],{},[119,4595,4596,4608,4620,4632,4644,4656,4668,4680],{},[107,4597,4598,4600,4602,4604,4606],{},[124,4599,686],{},[124,4601,1916],{},[124,4603,1919],{},[124,4605,1922],{},[124,4607,1925],{},[107,4609,4610,4612,4614,4616,4618],{},[124,4611,1930],{},[124,4613,1933],{},[124,4615,1936],{},[124,4617,753],{},[124,4619,1941],{},[107,4621,4622,4624,4626,4628,4630],{},[124,4623,1946],{},[124,4625,276],{},[124,4627,290],{},[124,4629,1953],{},[124,4631,290],{},[107,4633,4634,4636,4638,4640,4642],{},[124,4635,1031],{},[124,4637,1962],{},[124,4639,1965],{},[124,4641,276],{},[124,4643,276],{},[107,4645,4646,4648,4650,4652,4654],{},[124,4647,1974],{},[124,4649,276],{},[124,4651,1979],{},[124,4653,798],{},[124,4655,798],{},[107,4657,4658,4660,4662,4664,4666],{},[124,4659,1988],{},[124,4661,724],{},[124,4663,724],{},[124,4665,276],{},[124,4667,246],{},[107,4669,4670,4672,4674,4676,4678],{},[124,4671,1241],{},[124,4673,782],{},[124,4675,322],{},[124,4677,322],{},[124,4679,322],{},[107,4681,4682,4684,4686,4688,4690],{},[124,4683,2013],{},[124,4685,2016],{},[124,4687,2019],{},[124,4689,246],{},[124,4691,243],{},[20,4693,328],{"id":328},[40,4695,4696,4700,4704,4708,4712],{},[43,4697,4698,2033],{},[29,4699,2032],{},[43,4701,4702,2039],{},[29,4703,2038],{},[43,4705,4706,2045],{},[29,4707,2044],{},[43,4709,4710,2051],{},[29,4711,2050],{},[43,4713,4714,2057],{},[29,4715,2056],{},[20,4717,364],{"id":363},[40,4719,4720,4722,4724,4726,4728,4730,4732],{},[43,4721,2064],{},[43,4723,2067],{},[43,4725,2070],{},[43,4727,2073],{},[43,4729,2076],{},[43,4731,2079],{},[43,4733,2082],{},[20,4735,412],{"id":412},[40,4737,4738,4742,4746,4750],{},[43,4739,4740],{},[416,4741,882],{"href":881},[43,4743,4744],{},[416,4745,2096],{"href":2095},[43,4747,4748],{},[416,4749,2101],{"href":944},[43,4751,4752],{},[416,4753,2107],{"href":2106},[20,4755,431],{"id":431},[191,4757,4758,4763,4768],{},[43,4759,2114,4760],{},[416,4761,2117],{"href":2117,"rel":4762},[445],[43,4764,2121,4765],{},[416,4766,2124],{"href":2124,"rel":4767},[445],[43,4769,2128,4770],{},[416,4771,2131],{"href":2131,"rel":4772},[445],{"title":455,"searchDepth":456,"depth":456,"links":4774},[4775,4776,4777,4778,4779,4780,4781,4782,4783,4784],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":1817,"depth":459,"text":1818},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},[4786,4787,4788,4789],{"q":2149,"a":2150},{"q":2152,"a":2153},{"q":2155,"a":2156},{"q":2158,"a":2159},[476,942],{},[483,484,485],[4794,4795],{"plan":1802,"price":949,"features":2165,"notes":2166},{"plan":1808,"price":2168,"features":2169,"notes":2170},[959,2173],{"power":489,"ux":490,"price":490,"cn_support":456,"stability":489},{"title":227,"description":2146},[4800,4801,4802],{"name":2179,"url":2117,"accessed":967},{"name":2181,"url":2124,"accessed":967},{"name":2183,"url":2131,"accessed":967},[470,2187,502,2188,2189,2190,2191],{"id":2683,"title":230,"alternatives":4805,"api_compatible":15,"body":4806,"category":470,"chinese_friendly":456,"cover":3102,"description":3103,"domestic":473,"extension":474,"faq":5137,"free":473,"github":15,"languages":5142,"lastVerified":15,"meta":5143,"models":15,"navigation":479,"notSuitable":15,"opensource":479,"path":881,"pillar":481,"platforms":5144,"priceTable":5145,"pricing":3124,"published":957,"relatedPlaybooks":5147,"relatedReviews":15,"score":5148,"self_host":479,"seo":5149,"seoTitle":3128,"slug":14,"sources":5150,"stem":3135,"suitable":15,"tagline":3136,"tags":5154,"updated":967,"verdict":3142,"website":3143,"__hash__":3144},[12,511,13,510],{"type":17,"value":4807,"toc":5125},[4808,4810,4814,4816,4818,4866,4868,4870,4872,4876,4890,4894,4914,4916,4940,4942,5044,5046,5068,5070,5086,5088,5106,5108],[20,4809,23],{"id":22},[25,4811,2691,4812,2695],{},[1152,4813,2694],{},[25,4815,2698],{},[20,4817,38],{"id":38},[40,4819,4820,4824,4830,4834,4840,4850,4854,4858,4862],{},[43,4821,4822,2708],{},[29,4823,2707],{},[43,4825,4826,1762,4828],{},[29,4827,2713],{},[1152,4829,2716],{},[43,4831,4832,2722],{},[29,4833,2721],{},[43,4835,4836,1762,4838],{},[29,4837,1031],{},[1152,4839,2729],{},[43,4841,4842,1762,4844,2738,4846,2738,4848],{},[29,4843,2734],{},[1152,4845,2737],{},[1152,4847,2741],{},[1152,4849,2744],{},[43,4851,4852,2750],{},[29,4853,2749],{},[43,4855,4856,2756],{},[29,4857,2755],{},[43,4859,4860,2762],{},[29,4861,2761],{},[43,4863,4864,2768],{},[29,4865,2767],{},[20,4867,93],{"id":93},[25,4869,2773],{},[20,4871,2777],{"id":2776},[25,4873,4874],{},[29,4875,147],{},[40,4877,4878,4882,4884,4886,4888],{},[43,4879,4880,2789],{},[1152,4881,2788],{},[43,4883,2792],{},[43,4885,2795],{},[43,4887,2798],{},[43,4889,2801],{},[25,4891,4892],{},[29,4893,169],{},[40,4895,4896,4902,4906,4910,4912],{},[43,4897,2810,4898,2814,4900],{},[1152,4899,2813],{},[1152,4901,2817],{},[43,4903,2820,4904],{},[1152,4905,2823],{},[43,4907,2826,4908,2830],{},[1152,4909,2829],{},[43,4911,2833],{},[43,4913,2836],{},[20,4915,189],{"id":189},[191,4917,4918,4922,4926,4930,4934,4938],{},[43,4919,4920,2846],{},[1152,4921,2845],{},[43,4923,4924,2852],{},[1152,4925,2851],{},[43,4927,4928,2857],{},[1152,4929,2788],{},[43,4931,2860,4932],{},[1152,4933,2863],{},[43,4935,2866,4936],{},[1152,4937,2869],{},[43,4939,2872],{},[20,4941,213],{"id":213},[101,4943,4944,4958],{},[104,4945,4946],{},[107,4947,4948,4950,4952,4954,4956],{},[110,4949,222],{},[110,4951,230],{},[110,4953,227],{},[110,4955,424],{},[110,4957,1907],{},[119,4959,4960,4972,4984,4996,5008,5020,5032],{},[107,4961,4962,4964,4966,4968,4970],{},[124,4963,686],{},[124,4965,2899],{},[124,4967,2902],{},[124,4969,2905],{},[124,4971,2908],{},[107,4973,4974,4976,4978,4980,4982],{},[124,4975,189],{},[124,4977,2016],{},[124,4979,2016],{},[124,4981,246],{},[124,4983,243],{},[107,4985,4986,4988,4990,4992,4994],{},[124,4987,1930],{},[124,4989,2713],{},[124,4991,2929],{},[124,4993,753],{},[124,4995,753],{},[107,4997,4998,5000,5002,5004,5006],{},[124,4999,2938],{},[124,5001,1965],{},[124,5003,1962],{},[124,5005,276],{},[124,5007,276],{},[107,5009,5010,5012,5014,5016,5018],{},[124,5011,2951],{},[124,5013,2954],{},[124,5015,2957],{},[124,5017,2960],{},[124,5019,246],{},[107,5021,5022,5024,5026,5028,5030],{},[124,5023,1974],{},[124,5025,1979],{},[124,5027,276],{},[124,5029,1953],{},[124,5031,798],{},[107,5033,5034,5036,5038,5040,5042],{},[124,5035,1241],{},[124,5037,322],{},[124,5039,319],{},[124,5041,2985],{},[124,5043,322],{},[20,5045,328],{"id":328},[40,5047,5048,5052,5056,5060,5064],{},[43,5049,5050,2997],{},[29,5051,2996],{},[43,5053,5054,3003],{},[29,5055,3002],{},[43,5057,5058,3009],{},[29,5059,3008],{},[43,5061,5062,3015],{},[29,5063,3014],{},[43,5065,5066,3021],{},[29,5067,3020],{},[20,5069,364],{"id":363},[40,5071,5072,5074,5076,5078,5080,5082,5084],{},[43,5073,3028],{},[43,5075,3031],{},[43,5077,3034],{},[43,5079,3037],{},[43,5081,3040],{},[43,5083,3043],{},[43,5085,3046],{},[20,5087,412],{"id":412},[40,5089,5090,5094,5098,5102],{},[43,5091,5092],{},[416,5093,876],{"href":875},[43,5095,5096],{},[416,5097,2096],{"href":2095},[43,5099,5100],{},[416,5101,2101],{"href":944},[43,5103,5104],{},[416,5105,888],{"href":887},[20,5107,431],{"id":431},[191,5109,5110,5115,5120],{},[43,5111,3073,5112],{},[416,5113,3076],{"href":3076,"rel":5114},[445],[43,5116,3080,5117],{},[416,5118,3083],{"href":3083,"rel":5119},[445],[43,5121,2128,5122],{},[416,5123,2131],{"href":2131,"rel":5124},[445],{"title":455,"searchDepth":456,"depth":456,"links":5126},[5127,5128,5129,5130,5131,5132,5133,5134,5135,5136],{"id":22,"depth":459,"text":23},{"id":38,"depth":459,"text":38},{"id":93,"depth":459,"text":93},{"id":2776,"depth":459,"text":2777},{"id":189,"depth":459,"text":189},{"id":213,"depth":459,"text":213},{"id":328,"depth":459,"text":328},{"id":363,"depth":459,"text":364},{"id":412,"depth":459,"text":412},{"id":431,"depth":459,"text":431},[5138,5139,5140,5141],{"q":3106,"a":3107},{"q":3109,"a":3110},{"q":3112,"a":3113},{"q":3115,"a":3116},[476],{},[483,484,485,2653],[5146],{"plan":126,"price":949,"features":3122,"notes":3123},[959,2173],{"power":489,"ux":489,"price":490,"cn_support":456,"stability":490},{"title":230,"description":3103},[5151,5152,5153],{"name":3131,"url":3076,"accessed":967},{"name":3133,"url":3083,"accessed":967},{"name":2183,"url":2131,"accessed":967},[470,3138,3139,3140,3141,502,2188,2191,978],1785428438832]